Active suppression method for electromagnetic compatibility interference of new energy switching power supply

Abstract

The present application relates to the technical field of electromagnetic compatibility of switching power supply, in particular to a new energy switching power supply electromagnetic compatibility interference active suppression method, which synchronously collects conducted interference with switching clock as coherent reference, extracts complex vector of each switching harmonic component and identifies parasitic resonance peak frequency point; for target harmonic number, determines equal amplitude and opposite phase sine component of complex vector and complex gain of injection path at the frequency point, injects into power input terminal after summation; determines dead time between two switching tubes of the same bridge arm according to peak frequency point, so that the damping ring excited by two jumps is opposite and superimposed at the frequency point; feeds forward each complex vector with equivalent duty cycle change caused by dead time change; iteratively updates each frequency point below the limit value, degrades broadband cancellation to single frequency point vector zero, eliminates the restriction of injection path gain bandwidth product on cancellation depth, and weakens resonance excitation with the existing dead time, so that the minimum margin of full frequency band is improved by 18.1dB.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic compatibility technology for switching power supplies, and specifically to an active suppression method for electromagnetic compatibility interference in new energy switching power supplies. Background Technology

[0002] After replacing traditional silicon devices with wide-bandgap devices such as silicon carbide metal-oxide-semiconductor field-effect transistors (MOSFETs) and gallium nitride high-electron-mobility transistors (GaN transistors), the voltage change rate of switching transitions in new energy switching power supplies has increased from several volts per nanosecond to tens of volts per nanosecond. Power density and conversion efficiency have improved accordingly, but the conducted electromagnetic interference (EMI) generated by these switching transitions has also intensified. Whether conducted interference can meet the corresponding electromagnetic compatibility limits across the entire frequency band from 150kHz to 30MHz has become a decisive factor in whether new energy switching power supplies can obtain market access.

[0003] Existing suppression methods can be categorized into three types. The first type is passive filtering, which involves connecting a common-mode choke in series at the power input and bridging X and Y capacitors to form a two- or even three-stage attenuation network. The second type is active electromagnetic interference filtering, which involves setting up sensing and injection branches on the power conductor. After detecting the interference voltage or current through a broadband path composed of operational amplifiers, an inverted cancellation signal is injected back into the power conductor. Representative schemes are found in Chinese invention patent CN108696117B and US patent US6775157B2. The third type is source-side modulation, which involves processing the switching control signal itself, including switching frequency jitter spread spectrum and cycle-by-cycle deterministic frequency conversion. Representative schemes are found in US patent US8963527B2 and international application WO2001020822A1; as well as slow-varying gate drive to reduce gate drive strength and slow down switching transitions. A predictive active filtering scheme that has emerged in recent years is found in US Patent No. 11621628B2, which uses a discrete-time model to predict ripple online and feed forward to synthesize the cancellation signal, which is then fed into the line filter via an injection transformer.

[0004] All three of the above methods encounter unavoidable physical limitations in new energy switching power supplies.

[0005] For passive filters, the parasitic capacitance between the windings of the common-mode choke and the equivalent series inductance of the Y capacitor form a bypass branch above several megahertz, causing the actual attenuation capability of the filter to drop significantly from 6MHz to 10MHz. Meanwhile, the switching transition time of wide-bandgap devices has been shortened to the order of 20ns, and their first spectral cutoff frequency has shifted upwards to above 15MHz, with interference energy concentrated precisely in the frequency band where passive filters fail. Further increasing the size of magnetic components not only contradicts the power density target of new energy switching power supplies, but also increases their parasitic parameters with increasing size, leading to saturation of benefits.

[0006] For active electromagnetic interference (EMI) filtering, the suppression depth is limited by the gain-bandwidth product of the injection path. This type of scheme is essentially a broadband negative feedback or broadband feedforward loop. To ensure loop stability across the entire operating frequency band, the loop gain must decrease with increasing frequency. Therefore, attenuation of around 20dB can be achieved in the 150kHz to 2MHz range, while the attenuation capability approaches zero above 8MHz. More critically, the complex gain of the injection path depends on the equivalent impedance of the power input. However, the equivalent impedance of vehicle wiring harnesses, industrial distribution cabinets, and consumer electronics power supply environments differs significantly from the impedance stability network in laboratory settings. This leads to a substantial mismatch in the cancellation parameters set in the laboratory after actual vehicle or equipment installation, sometimes even resulting in a superposition of cancellations.

[0007] For source-side modulation, both switching frequency jitter spread spectrum and cycle-by-cycle deterministic frequency conversion are forms of energy redistribution. Their principle is to broaden the energy originally concentrated in a single spectral line to a range several times the measurement resolution bandwidth, thereby reducing receiver readings. This type of scheme is effective at higher harmonics because the absolute broadening of the spectral line is proportional to the harmonic order; however, at lower harmonics from 150kHz to 1MHz, the absolute broadening is less than the receiver's 9kHz resolution bandwidth, and the readings remain almost unchanged. While slowly varying gate drive can indeed reduce high-frequency interference, switching losses are proportional to the transition time. In a 3.3kW-class bidirectional DC-DC converter, slowing the transition time by 2.6 times increases the switching losses from 54W to 141W, and reduces the conversion efficiency by more than 2.4 percentage points. This is unacceptable in new energy switching power supplies where efficiency is a core performance indicator.

[0008] Upon further investigation, the reason why the above three methods are each limited lies in the contradiction between two underlying mechanisms that have not yet been addressed by existing technologies.

[0009] First, existing cancellation schemes treat conducted interference as broadband noise, necessitating the construction of a broadband path for anti-phase cancellation. This leads to a tradeoff between suppression bandwidth and loop stability margin, and the complex gain of the injection path becomes uncorrectable once it drifts with load impedance. However, conducted interference in switching power supplies is not broadband noise: it is a comb spectrum generated by the periodic repetition of switching transitions, strictly phase-locked with the switching clock, with the amplitude and phase of each spectral line being deterministic in steady state. Existing electromagnetic interference receivers and sampling circuits based on root-mean-square detection can only provide the amplitude of each spectral line, discarding its phase. This information loss prevents the coherent nature of the interference from being utilized, reducing cancellation to broadband tracking.

[0010] Secondly, new energy switching power supplies commonly exhibit underdamped resonances at tens of megahertz, formed by the parasitic inductance of the power circuit and the output capacitance of the switching transistor. This resonance is excited by each switching transition, forming a broad envelope peak superimposed on the comb spectrum. The energy of this peak comes from the energy stored in the parasitic loop injected instantaneously during the transition, and is directly related to the temporal structure of the transition, but unrelated to filtering along the propagation path. Therefore, both passive filtering and active injection on the power conductor can only attenuate this peak along the propagation path, but cannot weaken its excitation. While gradual gate driving can weaken the excitation, it comes at the cost of switching losses. In existing technologies, electromagnetic compatibility and conversion efficiency are directly opposed at this resonance peak, and there is currently no means to weaken the resonant excitation without increasing switching losses.

[0011] In summary, the core technical problems faced by existing technologies can be summarized as follows: conducted interference in switching power supplies manifests as a coherent comb spectrum phase-locked with the switching clock in the low-frequency band, and as a parasitic resonant envelope peak excited by abrupt transients in the high-frequency band. The physical causes, controllable variables, and cost functions of the two are completely different. Existing suppression methods all deal with the two by using a single broadband propagation path attenuation or a single source-end energy broadening. This results in the cancellation depth in the low-frequency band being limited by the injection path gain-bandwidth product, and the suppression of resonant peaks in the high-frequency band being directly opposed to the conversion efficiency. The electromagnetic compatibility margin and efficiency across the entire frequency band cannot be optimized simultaneously. Summary of the Invention

[0012] To address the aforementioned problems, this invention proposes an active suppression method for electromagnetic compatibility interference in new energy switching power supplies. The core concept of this invention is that since the low-frequency comb spectrum of conducted interference is phase-locked with the switching clock, the complex vector of each spectral line should be measured using the switching clock as a coherent reference. Then, a sinusoidal component with a fixed frequency, precise amplitude, and phase calculations should be synthesized for each target spectral line to cancel it out. Because each injected component is a steady-state sinusoidal quantity at a single frequency, the complex gain of the injection path at that single frequency point can be measured online and precisely pre-compensated. The cancellation depth is no longer constrained by the gain-bandwidth product and stability margin of the broadband loop, thus fundamentally bypassing the bandwidth bottleneck of existing active filters. Since the energy of the high-frequency resonant peak originates from the excitation of the parasitic loop by the switching transition, and the turn-off of the first switch and the turn-on of the second switch within the same bridge arm excite the same parasitic loop twice, with the time interval between the two excitations being precisely the dead time—an adjustable quantity that already exists and has a relatively wide and flat range near the efficiency optimum—the dead time should be tuned to a value that causes the two damped ringing strings to superimpose in opposite phase at the resonant frequency. One excitation is used to cancel out the other, thus weakening the resonant peak with almost no change in switching losses. The two suppression degrees of freedom correspond to two different physical causes and are not interchangeable. However, changing the dead time will cause a change in the equivalent duty cycle, which in turn changes the amplitude and phase of each component of the low-frequency comb spectrum. Therefore, this cross-coupling must be corrected using a feedforward method; otherwise, the two degrees of freedom will destroy each other.

[0013] The method of the present invention includes the following steps.

[0014] The first step is to synchronously acquire conducted interference signals using the switching clock of the switching power supply as a coherent reference, obtain the complex vectors of each switching harmonic component, and identify the peak frequency points in the interference spectrum formed by parasitic parameter resonance.

[0015] The second step involves determining the amplitude and phase of a sinusoidal component that is equal in amplitude and opposite in phase to the target harmonic number at the injection point by using the complex vector of the switching harmonic component and the complex gain of the injection path at the harmonic frequency. The sinusoidal components are then summed and injected into the input terminal of the switching power supply through the injection path.

[0016] The third step is to determine the dead time between the first switch being turned off and the second switch being turned on in the same bridge arm based on the peak frequency, so that the damping ringing excited by the first switch being turned off and the damping ringing excited by the second switch being turned on are superimposed in opposite phase at the peak frequency.

[0017] The fourth step is to use the equivalent duty cycle change caused by the change in dead time to feedforward and correct the complex vector of each switching harmonic component.

[0018] The fifth step involves iteratively updating the amplitude, phase, and dead time of each sinusoidal component based on the re-acquired complex vector until the interference amplitude at each frequency point is lower than the preset limit.

[0019] The beneficial effects of this invention are as follows.

[0020] First, synchronous accumulation using the switching clock as a coherent reference makes the phase of each switching harmonic component measurable. The coherent component increases linearly with the number of accumulation cycles, while the incoherent background increases with the square root of the number of accumulation cycles, thus improving the signal-to-noise ratio of the phase measurement. When the value is 64, the standard deviation of the phase measurement decreases from 11.4 degrees before accumulation to 1.4 degrees, providing the necessary accuracy basis for harmonic equal amplitude antiphase cancellation.

[0021] Second, harmonic-wise complex vector synchronous injection degrades broadband cancellation into vector zeroing of several single-frequency points. The complex gain of the injection path at each target frequency point is given by online identification and accurately pre-compensated. The cancellation depth of a single target spectral line reaches 26dB, and the peak level in the low-frequency band is reduced from 84.6dBμV to 58.6dBμV, an improvement of 26.0dB, which is significantly better than the 21.0dB improvement of line-injection active filtering under the same operating conditions. Moreover, this improvement does not decrease as the injection frequency point moves up.

[0022] Third, the dead-time resonant phase coordination weakens the resonant excitation by using the existing dead time as a control variable. Under the operating conditions of the on-board 3.3kW bidirectional DC-DC converter, the dead time is adjusted from 40.0ns (tuned according to efficiency) to 19.5ns, and the parasitic resonance peak is reduced from 68.8dBμV to 57.5dBμV. Meanwhile, the body diode freewheeling loss changes by only 0.42W, and the conversion efficiency changes from 97.190% to 97.124%, a decrease of only 0.067 percentage points. In contrast, the gradual gate drive achieves an improvement of 18.7dB in the resonance peak while reducing the efficiency by 2.429 percentage points.

[0023] Fourth, the tuning method, which uses the dead time and the amplitude ratio of the two damped ringing strings as two-dimensional optimization variables and aims to maximize the minimum margin across the entire frequency band, yielded a counterintuitive result: the optimal dead time is not the precise zero-alignment solution that maximizes the reduction of the resonance peak, and the optimal amplitude ratio is not the value of perfectly balanced two ringing strings, but rather falls within the intervals. In Example 1, although the precise zero-alignment solution reduces the resonance peak by 18.0 dB, it raises the frequency band by 12 MHz to 16 MHz, resulting in a minimum margin of only -4.6 dB across the entire frequency band; the two-dimensional optimization solution reduces the resonance peak by only 11.3 dB, but the minimum margin across the entire frequency band reaches +1.3 dB, a difference of 6.0 dB.

[0024] Fifth, there is a nonlinear synergy between the two suppression degrees of freedom. Applying harmonic-wise injection alone increases the minimum margin across the entire frequency band by 8.6 dB, while applying dead-zone phasing and amplitude optimization alone increases it by 0.0 dB. The combined effect of these two increases it by 18.1 dB, more than twice the sum of their individual effects. This is because the minimum margin across the entire frequency band is determined by the worst frequency point; only when the weaknesses in both the low-frequency and high-frequency bands are simultaneously addressed will the minimum margin truly shift upwards.

[0025] Sixth, the cross-coupling feedforward correction eliminates the potential for mutual destruction between the two degrees of freedom. A 1ns change in dead time at a 100kHz switching frequency causes a 0.01% change in the equivalent duty cycle, resulting in a 1.8-degree phase shift in the 100th harmonic. In Example 1, the dead time was adjusted from 40.0ns to 19.5ns. Without correction, the phase error of the 100th harmonic would reach 36.9 degrees, and its cancellation depth would degrade from 26dB to 4.0dB, meaning cancellation would almost completely fail. By applying the dead time simultaneously with the feedforward correction, the cancellation depth is restored, and the convergence lag caused by waiting for the next measurement is avoided.

[0026] Seventh, the method achieved compliance in all three application scenarios. In the automotive scenario, using the CISPR 25 Class 5 limit template, the minimum margin across the entire frequency band increased from -16.8dB to +1.3dB, and the number of out-of-compliance frequency points decreased from 128 to 0. In the industrial scenario, using the CISPR 11 Class B limit template, the minimum margin increased from -23.0dB to +3.0dB, and the number of out-of-compliance frequency points decreased from 75 to 0. In the consumer electronics scenario, using the CISPR 32 Class B limit template, the minimum margin increased from -14.9dB to +11.1dB, and the number of out-of-compliance frequency points decreased from 3 to 0. The closed-loop system achieved compliance from the 7th iteration, with a single iteration taking 0.62ms and a total convergence time of 4.9ms. Attached Figure Description

[0027] Figure 1 This is an overall flowchart of the method of the present invention.

[0028] Figure 2 This is a system architecture block diagram of a switching power supply and an active suppression unit for implementing the method of the present invention.

[0029] Figure 3 This is a timing diagram of synchronous accumulation and complex vector extraction based on a switching clock as the coherent reference.

[0030] Figure 4 This is a schematic diagram of the vector relationship for harmonic-wise complex vector synchronous injection.

[0031] Figure 5 It is a superimposed waveform diagram of two damped ringing rings with different dead times and different ringing amplitude ratios.

[0032] Figure 6 It is a two-dimensional minimum-maximum tuning curve of dead time versus ringing amplitude ratio.

[0033] Figure 7 This is a comparison diagram of the conducted interference spectrum and the limit line before and after implementing the method of the present invention.

[0034] Figure 8 This is a comparison diagram of the conducted interference spectrum of the method of the present invention and four control schemes.

[0035] Figure 9 It is a convergence curve of interference level versus minimum margin across the entire frequency band during the closed-loop iteration process.

[0036] Figure 10 This is a bar chart showing the minimum margin increase of each control protocol and each ablation protocol relative to the baseline. Detailed Implementation

[0037] The technical solution of the present invention will be further described below with reference to the accompanying drawings. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0038] Conducted interference in new energy switching power supplies originates from the switching node voltage. The periodic jumps. To The decomposition into two components, each with different physical causes, frequency domain morphology, and controllable variables, is the starting point for all the technical means of this invention.

[0039] The first component is an idealized periodic trapezoidal jump sequence. Let the switching period be... The switching frequency is Duty cycle is The rise time is The descent time is Bus voltage is Then the first of the components The Fourier coefficients are:

[0040] ,in express This equation shows that the trapezoidal jump sequence is located in the frequency domain. The amplitude of discrete spectral lines at integer multiples of the clock frequency is given by the product of two Sink's functions, and their phase is uniquely determined by the duty cycle and harmonic order. As long as the switching clock is stable, the complex vector of each spectral line is a deterministic quantity in steady state, rather than random noise. This is a key fact that existing cancellation schemes have not utilized.

[0041] The second component is the damped ringing generated by each switching transition's excitation of the parasitic loop. This is the parasitic inductance of the power circuit. Equivalent output capacitance of the switching transistor A series resonant loop is formed, and its resonant frequency and quality factor are: , ,in This is the equivalent damping resistance of the loop. Using the parameters of Example 1... Take 40nH, Take 1.4nF, Calculate using 3.2Ω. It is 21.27MHz. The value is 1.67, corresponding to a damping ratio of 0.299, a decay time constant of 25 ns, and a 3dB bandwidth of 12.73 MHz for the resonance peak. The energy of this ringing comes from the energy stored in the parasitic loop injected during the transition, which manifests in the frequency domain as a broad envelope peak superimposed on the comb spectrum.

[0042] The two components mentioned above form a common-mode displacement current path through the parasitic capacitance of the switching node to ground, and a differential-mode path through the power loop, both coupling to the power input terminal. After being attenuated by the existing passive electromagnetic interference filter, they reach the measurement port of the line impedance stabilization network. The existing passive filter attenuates the first component by more than 40dB below 3MHz, but due to the bypass effect of the capacitance between the common-mode choke windings and the equivalent series inductance of the Y capacitor, the attenuation capability drops back to around 6dB above 6MHz, just enough to allow the second component to pass.

[0043] Accordingly, the present invention sets two suppressed degrees of freedom with different physical causes and different cost functions.

[0044] The first degree of freedom pertains to the first component. Since each spectral line of this component is a definite complex vector, a sinusoidal component of the same frequency can be synthesized separately for each target spectral line, making it equal in amplitude and out of phase with the spectral line at the injection point. Each injected component is a steady-state sinusoidal quantity at a single frequency, therefore the complex gain of the injection path at that single frequency is a definite complex number, which can be measured online and precisely pre-compensated. The cancellation problem thus degenerates from constructing a broadband out-of-phase path to completing vector zeroing at several isolated frequency points, and the gain-bandwidth product of the injection path and the loop stability margin no longer constitute constraints. The cost of this degree of freedom is only the static power consumption of the injection power amplifier, and is independent of switching losses.

[0045] The second degree of freedom pertains to the second component. The turn-off of the first switch and the turn-on of the second switch in the same bridge arm will excite the same parasitic loop twice, and the time interval between these two events is the dead time. Let the magnitudes of the two incentives be respectively... and The difference in excitation phase is Then the coherent superposition factor of the two ringing strings at frequency f is: Let the factor be in Taking the minimum value, we obtain the inversion condition: There is a crucial mechanistic point here that must be clarified: the coherent superposition factor. The interference period on the frequency axis is The reciprocal of the value. Only when the interference period is much larger than the 3dB bandwidth of the resonance peak can an interference zero point cover the entire resonance peak, thus truly suppressing the envelope. Calculated using Example 1, When the interference period is 19.5 ns, the interference period is 51.3 MHz, while the 3 dB bandwidth of the resonance peak is 12.73 MHz. The former is 4.0 times that of the latter, thus satisfying the condition. As a counter-evidence, if the time interval between the on-edge and off-edge within the same switching cycle, i.e., the on-time, is used as the control variable, this interval is 4200 ns, corresponding to an interference period of only 238 kHz, which is much smaller than the resonance peak bandwidth. In this case, adjacent spectral lines alternately fall into the constructive and destructive interference points, and the envelope height of the resonance peak does not decrease; the energy is merely redistributed among the spectral lines. This difference determines that the dead time, rather than the on-time, must be used as the control variable for the high-frequency band. The cost of this degree of freedom is only the change in conduction loss caused by the change in the freewheeling time of the body diode, which is 0.42 W in Example 1, accounting for 0.013% of the rated output power.

[0046] There is a unidirectional cross-coupling chain between the two degrees of freedom. Changing the dead time will change the actual on-time of the switch, that is, change the equivalent duty cycle. As can be seen from the Fourier coefficient expression of the first component, a change in D will simultaneously alter the amplitude and phase of each spectral line, thus causing a mismatch in the injection vector already calculated for the first degree of freedom. Conversely, the injection component is applied to the power input terminal without changing the timing of the switching transistors, and therefore does not affect the second degree of freedom. Therefore, this invention uses a feedforward method to correct the complex vectors of each harmonic while applying the dead time, without waiting for the next round of measurement.

[0047] Example 1

[0048] like Figure 1 and Figure 2As shown, the switching power supply in this embodiment is an on-board bidirectional DC-DC converter, employing a synchronous buck topology. The first and second switching transistors form the same bridge arm, both being silicon carbide metal-oxide-semiconductor field-effect transistors. The bus voltage is 400V, the output voltage is 48V, the rated output current is 68A, the rated output power is 3300W, the switching frequency is 100kHz, the duty cycle is 0.42, the rise time is 22ns, the fall time is 18ns, and the dead time, tuned for efficiency, is 40ns. The input terminal is equipped with a two-stage passive electromagnetic interference filter. The electromagnetic compatibility limits are based on the CISPR 25 Class 5 conducted emission limit template, i.e., 80dBμV for 150kHz to 530kHz, 70dBμV for 530kHz to 2MHz, and 62dBμV for 2MHz to 30MHz.

[0049] The active suppression unit consists of an electromagnetic compatibility detection interface, an analog-to-digital converter, a digital signal processor, a digital-to-analog converter, an injection power amplifier, an injection transformer, and a segmented gate driver. The digital signal processor shares the same clock source with the original pulse width modulation controller of the switching power supply to ensure the strictness of the coherence reference.

[0050] Step S1: Synchronously acquire conducted interference signals and extract complex vectors using the switching clock as a coherent reference. The electromagnetic compatibility detection interface extracts the conducted interference signal from the power input terminal via a coupling network and sends it to a 12-bit analog-to-digital converter. The sampling trigger reference is taken from the conduction edge of the first switching transistor drive signal, meaning that the sampling start point for each switching cycle is locked at the same physical moment. The sampling rate is 1024 times the switching frequency, i.e., 102.4 MSPS, resulting in 1024 sampling points per switching cycle, forming a single-cycle sampling sequence.

[0051] Continuous The single-cycle sampling sequence of each switching cycle is accumulated point by point according to the index within the cycle. Let the first... The first cycle The value of each sampling point is Then the coherent cumulative sequence is: The components coherent with the switching clock are in the same phase in each cycle, and their amplitudes, after accumulation, are... Linear growth; while incoherent components such as thermal noise, interference from other electrical equipment, and quantization noise are uncorrelated between periods, and their effective value after accumulation follows a linear growth pattern. The square root of the increase. Therefore, coherent accumulation improves the signal-to-noise ratio. The square root times. Taking 64 values ​​improves the signal-to-noise ratio by 8 times, or 18.1 dB. The standard deviation of the phase measurement is approximately equal to the reciprocal of the signal-to-noise ratio, which is 0.199 rad (11.4 degrees) without accumulation and decreases to 0.0249 rad (1.4 degrees) after accumulation. This accuracy is a prerequisite for subsequent equal-amplitude phase-reversal cancellation to achieve a depth of 26 dB: the cancellation depth is jointly limited by the relative amplitude error and the phase error, with an upper limit of... relative error of amplitude Take 0.01, phase error Using 0.0249 rad for calculation, the theoretical upper limit of the cancellation depth is 31.5 dB; after taking into account the quantization error of the digital-to-analog converter and the nonlinearity of the power amplifier, the stable cancellation depth achieved in engineering is 26 dB.

[0052] Perform a 1024-point Discrete Fourier Transform on the coherent cumulative sequence, and take the index as... The spectral line, whose complex value is the th... The complex vector of the second-order switching harmonic component is denoted as: , where the modulus The amplitude and argument of this harmonic are... This represents the phase of the harmonic. The value ranges from 16 to 256. When the value is less than 16, the signal-to-noise ratio improvement is less than 12dB, the phase standard deviation is greater than 3.5 degrees, and the cancellation depth is less than 20dB. When the value is greater than 256, the single-round measurement time exceeds 2.56ms, the closed-loop bandwidth is lower than the response requirements of the load step, and the drift of the switching power supply operating point within the measurement window will introduce new phase errors. The time taken for a single-wheel measurement is 0.64ms when the value is 64, which is a value that balances accuracy and response speed. Figure 3 The timing relationship between the synchronous accumulation and the complex vector extraction is shown.

[0053] Parasitic resonance peaks are identified on the amplitude spectrum. A peak-holding envelope with a 13% relative bandwidth is taken for each harmonic amplitude. Local maxima within the 3MHz to 30MHz range are searched on this envelope; the corresponding frequencies are the peak frequencies. The frequency interval at which the frequency drops by 3 dB on both sides is the resonant peak bandwidth, and the ratio of the two is the quality factor. The estimated value. This embodiment identifies the obtained... It has a frequency of 21.27MHz and a bandwidth of 12.73MHz. It is 1.67.

[0054] Step S2: Online identification of the complex gain of the injection path. The injection path extends from the digital-to-analog converter output through the power amplifier and injection transformer to the power input terminal, and its complex gain... It depends on the frequency response of the power amplifier, the leakage inductance and inter-turn capacitance of the injection transformer, the impedance of the coupling capacitor, and the equivalent impedance of the power input terminal. The equivalent impedance of the power input terminal varies with the length of the vehicle wiring harness, the battery status, and other parallel loads. The main source of drift. If the fixed parameters calibrated in the laboratory on the line impedance stabilization network are used, after actual installation... The amplitude deviation can reach 6dB and the phase deviation can reach 40 degrees. The cancellation depth will degrade to less than 3dB, and even when the phase deviation exceeds 90 degrees, it will turn from cancellation to superposition.

[0055] This invention replaces fixed calibration with online identification. For the first... For each target harmonic frequency, first record the current complex vector. Subsequently, a trial component with known amplitude and phase is applied to the injection pathway. Its amplitude is taken The amplitude is 0.25 to 0.33 times that of the amplitude, and the phase is arbitrary; the result is obtained by measurement again. The change in response is:

[0056] The complex gain of the injection path at this frequency is: A compromise was reached, with the amplitude ranging from 0.25 to 0.33 times the standard value: too small an amplitude would result in the response change being submerged in measurement noise, while too large an amplitude might cause the frequency point to briefly exceed the limit during identification. The 24 target harmonic frequencies were traversed one by one, with each frequency point occupying two measurement windows. The total identification time was 24 x 2 x 0.64 ms, or 30.7 ms. The amplitude repeatability of the identified complex gain was ±0.4 dB, and the phase repeatability was ±2.1 degrees.

[0057] While online identification is accurate, its 30.7ms execution time is unsuitable for repeated execution when the load or line impedance changes rapidly. Therefore, a feedforward neural network is used to quickly extrapolate the complex gain. This network is a three-layer feedforward structure. The input layer is a 5-dimensional feature vector, consisting of the waveform distortion rate of the input current within the power frequency cycle, the input equivalent impedance surrogate, the output current of the switching power supply, the case temperature of the switching transistor, and the root mean square value of the residual amplitude at each target harmonic order. The input equivalent impedance surrogate is taken as the low-frequency component of the ratio of the output voltage to the output current injected into the power amplifier; this value can be directly obtained from the existing power amplifier sampling circuit without additional hardware. The hidden layer has 12 neurons, with the hyperbolic tangent function as the activation function. The output layer is 48-dimensional, representing the complex gain amplitude correction and phase correction for each of the 24 target frequency points. The loss function is the sum of the squares of the residual amplitudes at each target harmonic order, and the weights are updated online using a backpropagation algorithm during closed-loop operation, with a learning rate of 0.002. When the equivalent impedance surrogate value at the input terminal changes by more than 15% compared to the value during the previous online identification, the complex gain is directly corrected using the correction value output by the network, without re-performing online identification. The correction time is less than 0.05ms. After 40 hours of full-vehicle road spectrum operation, the complex gain extrapolation error of this network under impedance abrupt change conditions is 0.9dB in amplitude and 4.6 degrees in phase, corresponding to a cancellation depth of 20.4dB. This represents a significant improvement compared to the less than 3dB error without correction, while also avoiding a 30.7ms identification pause.

[0058] Step S3: Harmonic-wise complex vector synchronous injection. The selection of target harmonic orders is based on margin criteria. Calculate the difference between the interference amplitude and the limit at each frequency point, and select the harmonic orders corresponding to the frequency points with the smallest difference (i.e., the worst margin) as target harmonic orders. The number of these is limited by the sine synthesis resources of the digital signal processor and the power capability of the injection power amplifier; in this embodiment, 24 are selected. Since the gain of the injection path drops sharply above 10MHz due to the leakage inductance of the injection transformer, the target harmonic orders are limited to within 100, i.e., within 10MHz. The target harmonic orders selected in this embodiment are 1, 2, 3, 4, 6, 56, 61, 63, 66, 68, 70, 73, 75, 78, 80, 82, 85, 87, 90, 92, 94, 95, 97, and 99.

[0059] For each target harmonic order k, the required injection component is: This means dividing the complex vector of the harmonic by the complex gain of the injection path at that frequency, and then taking the negative sign. The physical meaning of this formula is: to ensure that the injected component, after passing through the path, presents at the injection point as a quantity with the same amplitude but opposite phase to the interfering harmonic, the inverse of the path's complex gain must be pre-applied in the digital domain; that is, the amplitude must be pre-divided by the path gain, and the phase must be pre-subtracted from the path phase shift. This is the entirety of precise pre-compensation for single-frequency complex gain, and its feasibility depends entirely on the premise that each injected component is a steady-state sinusoidal quantity of a single frequency. Figure 4 The vector relationship is shown: the interference harmonic complex vector and the injected component are opposite in direction and equal in length at the injection point. After adding the two, only the residuals corresponding to the measurement and synthesis errors remain.

[0060] The 24 injected components are summed in the digital domain to obtain a synthesized waveform, which is then fed into the primary winding of the injection transformer via a digital-to-analog converter and a power amplifier. The secondary winding is connected in series with the positive terminal of the power input. The peak amplitude of the synthesized waveform needs to be limited; in this embodiment, the limiting value is set to 80% of the linear output range of the injection power amplifier. When the amplitude of the summed signal exceeds this limiting value, the amplitude is allocated according to the margin at each target harmonic order from smallest to largest. That is, priority is given to ensuring that the spectral line with the worst margin receives complete cancellation amplitude until the cumulative amplitude reaches the limiting value. The remaining spectral lines are not canceled in this round.

[0061] After injection, the low-frequency peak level decreased from 84.6 dBμV to 58.6 dBμV, an improvement of 26.0 dB. In contrast, under the same operating conditions, the improvement from line-injected active electromagnetic interference filtering was 21.0 dB, and this improvement decreased with increasing frequency, approaching zero above 8 MHz. The 26.0 dB improvement achieved by this method remained consistent across all target frequencies from 100 kHz to 10 MHz because compensation at each frequency was performed independently.

[0062] Step S4: Two-dimensional minimax tuning of dead-time resonant phase and ringing amplitude ratio. The peak frequency point identified in step S1... The frequency is 21.27MHz, and its period is 47.02ns. The first switch turning off causes a drop in the switching node voltage, and the second switch turning on also causes a drop in the switching node voltage. The excitation of the parasitic loop by both switches is in the same direction, hence the difference in excitation phase is... Approximately zero. and Substitute the inversion condition: The set of candidate dead time values ​​obtained are 23.51ns, 70.53ns, 117.55ns, etc. The preset allowable dead time range is 15ns to 90ns, with its lower limit determined by the maximum dispersion of the switch turn-off delay and the shoot-through risk, and its upper limit determined by the acceptable amount of freewheeling loss of the body diode. The value within this range that has the smallest difference from the current dead time of 40.0ns is 23.51ns; therefore, the precise alignment solution for the resonant zero is 23.51ns.

[0063] Based on this, a second adjustable variable is introduced. A single switching transition is divided into two driving segments with different gate drive currents: the first driving segment, from the rise of the gate-source voltage until reaching the Miller plateau, uses a larger gate drive current to quickly cross the high-voltage and high-current overlap region; the second driving segment, during the Miller plateau, has a gate drive current adjustable within the range of 0.35 to 1.0 times the current of the first driving segment. The current of the second driving segment determines the voltage change rate of this transition, and thus determines the energy stored in the parasitic loop injected during this transition. Therefore, adjusting the current of the second driving segment when the second switch is turned on can adjust the amplitude ratio of the second series of damped ringing relative to the first series of damped ringing. Divide by In this embodiment, the original value of the amplitude ratio is 0.62, and the adjustable range is 0.45 to 1.00. Since the second driving segment is located during the Miller plateau, the voltage across the switching transistor has not yet completely dropped. The impact of adjusting the current in this segment on the switching loss is much smaller than the overall slowing-down transition. In actual measurements, the switching loss increased by 1.9W when the amplitude ratio was adjusted from 0.62 to 0.76.

[0064] Dead zone time Compared with amplitude Divide by As a two-dimensional optimization variable, the objective is to maximize the minimum margin across the entire frequency band, using a 400x60 grid search within the intervals of 15ns to 90ns and 0.45 to 1.00. The results are counterintuitive and warrant detailed explanation.

[0065] Firstly, the optimal dead time is 19.51ns, while the non-resonant zero-point precise alignment solution is 23.51ns. With 23.51ns, the coherent superposition factor does indeed reach a minimum at 21.27MHz, resulting in an 18.0dB decrease in the resonant peak. However, the same coherent superposition factor is greater than 1 in the 12MHz to 16MHz frequency band, raising the frequency by 6.5dB. Since the interference level in this band was already close to the limit, this raise results in exceeding the limit, with a minimum margin of only -4.6dB across the entire frequency band and 46 exceeding frequency points. With 19.51ns, the resonant peak only decreases by 11.3dB, but the increase in the 12MHz to 16MHz frequency band is reduced to less than 1dB, achieving a minimum margin of +1.3dB across the entire frequency band and zero exceeding frequency points. The two solutions differ by 4.0ns, and the minimum margin across the entire frequency band differs by 6.0dB. This indicates that the absolute decrease in the resonant peak is not a meaningful optimization target; the minimum margin across the entire frequency band is. Precisely aligning the interference zero point with the resonant peak is equivalent to pushing the constructive interference point to one side of the resonant peak's skirt, and the skirt frequency band is often where the margin is most strained.

[0066] Secondly, the optimal amplitude ratio is 0.758, which is neither the original mismatch value of 0.62 nor 0.96, which makes the two ringing strings nearly perfectly balanced. The closer the amplitude ratio is to 1, the deeper and narrower the interference null point; the smaller the amplitude ratio, the shallower and wider the interference null point. When the dead time has deviated from the resonant null point, what is needed is not a deep and narrow null point, but a gentle dip covering the entire resonant peak and its skirt. Therefore, the optimal amplitude ratio falls within the range. Numerically, the minimum margin across the entire frequency band is +0.93dB when the amplitude ratio is 0.62, +1.02dB when the amplitude ratio is 0.96, and +1.33dB when the amplitude ratio is 0.758.

[0067] Figure 5 The time-domain superposition waveforms of two damped ringing strings are shown for three different values. Figure 6 The curves showing the minimum margin versus dead time for three amplitude ratio values, as well as the location of the two-dimensional minima and maxima solutions, are presented.

[0068] After the dead time is adjusted from 40.0 ns to 19.51 ns, the freewheeling time of the body diode is shortened, and its freewheeling loss changes according to the following formula: With the forward voltage drop of the body diode Calculated using 1.5V, 68A output current, and 100kHz switching frequency, the result is -0.42W, meaning the freewheeling loss actually decreases. After factoring in the 1.9W increase in switching loss due to the segmented gate drive adjustment ratio and the 0.55W static power consumption of the active suppression unit itself, the total loss increases from 95.4W to 97.7W, and the conversion efficiency decreases from 97.190% to 97.124%, a drop of 0.067 percentage points.

[0069] Step S5: Cross-coupling feedforward correction. Dead time from Become Subsequently, the actual on-time of the first switching transistor changes accordingly, and the change in the equivalent duty cycle is: In this embodiment, this quantity is -20.49 ns divided by 10 μs, which is -0.205%. From the Fourier coefficient expression described in step one, it can be seen that the amplitude and phase of each harmonic complex vector change with the duty cycle: the amplitude is scaled according to the ratio of the Sink function values ​​at the equivalent duty cycle before and after correction, i.e.: The phase is shifted by the circumferential angle of the product of the harmonic order and the change in the equivalent duty cycle, i.e.: For the 100th harmonic, the phase shift is π multiplied by 100 multiplied by -0.00205, which is 0.644 rad, or 36.9 degrees. Without correction, there is a 36.9-degree phase mismatch between the injected component of this harmonic and the actual interference. Its residual amplitude is twice the original amplitude (18.45 degrees, or 0.633 times), and the cancellation depth degrades from 26 dB to 4.0 dB, meaning the cancellation is almost completely ineffective. Regarding amplitude, k multiplied by... The change is 0.205, which is one-fifth of the width of one side lobe of the Sink function, and the magnitude ratio deviates from 1 by more than 20%.

[0070] This invention applies the aforementioned amplitude scaling and phase shift simultaneously with the application of a new dead time in the current iteration, and then recalculates each injected component based on the corrected complex vector. The reason for using feedforward instead of waiting for the next measurement before closed-loop correction is that the dead time adjustment can reach tens of nanoseconds during the optimization process. If the next measurement is waited for, the injected components in this round will be in a severely mismatched state throughout the entire measurement window. Not only will the interference level in this round spike briefly, but the residual after the mismatch will mislead the judgment of minimax redistribution, causing the iteration to oscillate between two incorrect operating points. After adding feedforward correction, the number of iterations required for measured convergence is reduced from 23 to 7.

[0071] Step S6: Minimal-Maximum Resource Reallocation and Closed-Loop Iteration. The difference between the interference amplitude at each frequency point and a preset limit is defined as the margin for that frequency point. The smallest margin among all frequency points is taken as the evaluation metric. The closed-loop iteration aims to maximize the evaluation metric, rather than maximizing the cancellation depth of any single frequency point.

[0072] During the iteration process, a phenomenon known as "balloon squeezing" occurs: the margin at a certain target harmonic order increases due to cancellation, but the margin at other frequencies decreases, and the evaluation value does not improve. There are two reasons for this: first, when the injected components are synthesized by the power amplifier, the intermodulation products between the components fall on non-target frequencies; second, when the injected components flow through the power input terminal, they change the equivalent impedance at that point, altering the interference transmission path at non-target frequencies. In this case, if the cancellation depth of the target harmonic order is further increased, the evaluation value will not only fail to increase but will actually decrease.

[0073] The processing rule of this invention is as follows: when the margin at the target harmonic number increases, the margin at other frequency points decreases, and the evaluation value does not increase in the current iteration, the target harmonic number whose margin is already greater than a preset threshold (6dB in this embodiment) is removed from the target harmonic number set, and the harmonic number corresponding to the frequency point where the evaluation value is located is moved into the set. The essence of this rule is to continuously allocate limited injection resources to the shortest current board, rather than further deepening the cancellation at already sufficiently generous frequency points. The maximization of the evaluation value is used as the iteration termination criterion, that is, the iteration terminates when the increment of the evaluation value is less than 0.2dB for three consecutive iterations.

[0074] In addition, two protection mechanisms are implemented. First, if the residual amplitude at each target harmonic order does not decrease in two consecutive iterations, it is determined that the complex gain is mismatched. The amplitude and phase of each injected component are then rolled back to the values ​​of the previous iteration, and the complex gain correction in step S2 is triggered. Second, the on-state voltage drop of the switching transistor is used as the junction temperature surcharge. When the junction temperature surcharge exceeds a preset threshold (corresponding to a junction temperature of 145 degrees Celsius), the dead time update is frozen and the current value is maintained to prevent the change in follow current loss caused by dead time adjustment from further increasing the junction temperature under high-temperature operating conditions.

[0075] The closed-loop convergence process of this embodiment is shown in Table 1.

[0076] Table 1 Closed-loop iterative convergence process

[0077] 0 84.6 68.8 -16.8 0.58 1 71.9 66.0 -7.6 1.20 2 65.4 62.0 -6.4 1.82 3 62.1 57.8 -4.8 2.44 4 60.4 54.9 -3.2 3.06 5 59.5 54.6 -1.8 3.68 6 59.0 55.3 -0.8 4.30 7 58.8 56.1 0.0 4.92 8 58.7 56.6 0.5 5.54 10 58.6 57.1 1.0 6.78 12 58.6 57.3 1.3 8.02 14 58.6 57.4 1.4 9.26

[0078] As shown in Table 1, the minimum margin across the entire frequency band turned positive starting from the 7th iteration, with a cumulative time of 4.92ms; it reached 1.4dB by the 14th iteration and tended to stabilize, with a total time of 9.26ms. The resonant peak level reached its lowest point of 54.6dBμV in the 5th iteration and then rebounded to 57.4dBμV. This is precisely the effect of minimax redistribution: the iteration transferred the excess margin on the resonant peak to the 12MHz to 16MHz frequency band, which had a worse margin at the time, causing the evaluation value to continue to rise. Figure 9 The convergence curve is shown.

[0079] Example 2

[0080] This embodiment of the switching power supply is an industrial totem-pole bridgeless power factor correction power supply. Two silicon carbide metal-oxide-semiconductor field-effect transistors (MOSFETs) in the high-frequency bridge arm constitute the same bridge arm. The output voltage is 400V, the rated output current is 24A, the rated output power is 5500W, the switching frequency is 65kHz, the duty cycle is 0.48, the rise time is 30ns, the fall time is 26ns, and the dead time, tuned for efficiency, is 55ns. The power loop parasitic inductance is 55nH, the equivalent output capacitance is 2.2nF, the switching node-to-ground parasitic capacitance is 165pF, and the loop equivalent damping resistance is 4.1Ω. Electromagnetic compatibility limits are based on the CISPR 11 Class B conducted emission limit template.

[0081] The peak frequency identified in step S1 is 14.47MHz. Following step S4 to solve for the phase inversion condition, the candidate dead time values ​​are 34.56ns, 103.68ns, etc. Due to the limitation of the lower limit of the allowable dead time range for industrial power supplies caused by the discreteness of turn-off delay due to larger chip area, it is set to 20ns, and the upper limit is set to 90ns. The optimal dead time obtained by two-dimensional minimax tuning is 22.14ns, and the optimal amplitude ratio is 1.000, meaning that in this embodiment, the optimal amplitude ratio falls within the upper limit of the range. This differs from the result of embodiment 1 because the 3dB bandwidth of the resonance peak in this embodiment is 8.9MHz, while the interference period is 45.2MHz, resulting in a ratio of 5.1, which is greater than the 4.0 in embodiment 1. The interference zero point is relatively wider and gentler than the resonance peak, thus allowing for a deeper and narrower zero point. This difference indicates that the amplitude ratio cannot be fixed based on experience and must be optimized online according to the actual identified resonance parameters.

[0082] After adjusting the dead time from 55ns to 22.14ns, the body diode freewheeling loss changed by -0.15W. The low-frequency peak level decreased from 75.8dBμV to 49.8dBμV, an improvement of 26.0dB; the parasitic resonance peak decreased from 49.5dBμV to 29.6dBμV, an improvement of 19.9dB; the minimum margin across the entire frequency band increased from -23.0dB to +3.0dB, and the number of out-of-range frequencies decreased from 75 to 0.

[0083] Based on the scenario-differentiated rules, the frequency band weights for industrial scenarios are tightened according to the characteristics of CISPR 11 Class B in the 500kHz to 5MHz range. This range is assigned a weight of 1.4, while the other frequency bands are assigned a weight of 1.0, so that the selection of target harmonic orders is tilted towards this range.

[0084] Example 3

[0085] The switching power supply in this embodiment is an active clamp flyback adapter for consumer electronics. The main switching transistor and the clamping switching transistor form the same bridge arm, both being gallium nitride high electron mobility transistors. It has an output power of 240W, an output current of 12.5A, a switching frequency of 140kHz, a duty cycle of 0.38, a rise time of 9ns, a fall time of 7.5ns, and a dead time of 22ns tuned for efficiency. The power circuit parasitic inductance is 18nH, and the equivalent output capacitance is 0.32nF. Electromagnetic compatibility limits are based on the CISPR 32 Class B conducted emission limit template, with a frequency upper limit of 30MHz.

[0086] The peak frequency identified in step S1 is 66.31MHz, which is higher than the upper limit of 30MHz for the CISPR32 conducted transmission frequency band. The rule at this point is to keep the dead time unchanged at 22ns (tuned for efficiency), not to perform dead-time phase matching in step S4, and to increase the number of target harmonic orders to the upper limit allowed by the digital signal processor. The technical basis for this is that when the peak frequency is outside the measurement frequency band, the reduction in resonance peak caused by dead-time phase matching is not included in the compliance judgment, while the increase in interference factor within the measurement frequency band is included. Therefore, performing dead-time phase matching in this case only reduces the evaluation value. This rule allows the method to automatically switch resource allocation methods in different scenarios without requiring separate tuning for each scenario.

[0087] While maintaining a dead time of 22ns, the low-frequency peak level decreased from 68.7dBμV to 42.7dBμV, an improvement of 26.0dB. The minimum margin across the entire frequency band increased from -14.9dB to +11.1dB, and the number of out-of-range frequencies decreased from 3 to 0. Since the margin bottleneck in this embodiment is entirely located in the low-frequency band, a positive margin of 11.1dB can be obtained by injecting only one degree of freedom through harmonic-wise complex vector synchronous injection, without needing to use a second degree of freedom. As a result, the power consumption of the active suppression unit is reduced to 0.31W.

[0088] Based on the operating conditions of Example 1 and the CISPR25 Class 5 limit template, four existing schemes and four ablation schemes were compared, and the results are shown in Table 2. The low-frequency peak was taken as the maximum level in the range of 150kHz to 3MHz, the high-frequency peak as the maximum level in the range of 3MHz to 30MHz, the resonant peak as the level at 21.27MHz, the minimum margin as the minimum value of the margin at each frequency point across the entire frequency band, and the frequency points exceeding the standard as the number of frequency points with negative margins.

[0089] Table 2 Performance Comparison of Each Solution

[0090]

[0091] Four conclusions can be drawn from Table 2.

[0092] First, none of the four existing solutions can achieve full-band compliance. Comparison 1, Comparison 3, and Comparison 4 all have a minimum margin of -16.8dB, with their bottleneck located in the low harmonics range of 150kHz to 530kHz. The absolute broadening of spread spectrum modulation at this point is less than the receiver's 9kHz resolution bandwidth, and the Sink envelope of the gradually varying gate drive remains almost unchanged, thus neither of these measures addresses this bottleneck. While Comparison 2 improves the low-frequency band by 21.0dB, its bandwidth limitation has no effect on the 21.27MHz resonant peak, shifting the bottleneck to the resonant peak, resulting in a minimum margin of -8.2dB.

[0093] Second, neither of the two degrees of freedom is sufficient on its own; their superposition exhibits a nonlinear synergy. Ablation A increases the minimum margin by 8.6 dB, while ablation C increases it by 0.0 dB. If linear superposition were expected, their combined effect should have increased it by 8.6 dB; however, the actual combined effect resulted in an 18.1 dB increase, 2.1 times the linear expectation. The mechanism lies in the fact that the minimum margin across the entire frequency band is determined by the worst-case frequency: ablation A raises the low-frequency band but shifts the bottleneck to the resonant peak; ablation C raises the resonant peak but the bottleneck remains in the low-frequency band. When either acts alone, they are locked by the other's bottleneck; only when applied simultaneously can the bottleneck truly shift upwards. This synergistic relationship cannot be deduced from either degree of freedom alone.

[0094] Third, the absolute decrease in the resonant peak is not positively correlated with full-band compliance. The resonant peak of ablation D is 50.8 dBμV, which is 6.7 dB lower than the 57.5 dBμV of this method, but its minimum margin is 5.9 dB lower than that of this method, and it has 46 out-of-standard frequency points, while this method has 0. This again confirms that the minimax criterion optimizes the weakest link, while maximizing the single-point cancellation depth will waste the margin on already ample frequency points and transfer the cost to the skirt band.

[0095] Fourth, the difference in efficiency cost is significant. This method achieves an 18.1 dB margin improvement with an efficiency cost of 0.067 percentage points, while the control group (comparison 4) achieves only a 0.0 dB margin improvement with an efficiency cost of 2.429 percentage points. The 36-fold difference in efficiency cost stems from the fact that the control group (comparison 4) slows down the entire switching transition, and its increased losses are proportional to the product of the transition time increment, bus voltage, and load current; while this method adjusts the dead time and the gate current during the Miller plateau. The former only changes the freewheeling time of the body diode, and the latter only affects the section where the voltage has not yet fully dropped.

[0096] Figure 7 The conducted interference spectrum and limit lines before and after implementing this method are shown. Figure 8 The spectral density of this method is compared with that of four control schemes. Figure 10 This shows the minimum margin improvement of each option relative to the baseline.

[0097] The technical effects of this invention can be attributed to three causal chains.

[0098] The first causal chain begins with coherence. The fact that conducted interference from the switching power supply in the low-frequency band is a coherent comb spectrum phase-locked with the switching clock makes the phase of each harmonic measurable. Synchronous accumulation, triggered by the switching clock, reduces the standard deviation of the phase measurement from 11.4 degrees to 1.4 degrees. Phase measurability transforms cancellation from broadband tracking to single-frequency vector zeroing. Single-frequency cancellation makes the complex gain of the injection path a identifiable and pre-compensated definite complex number, thus completely eliminating the dual constraints of the gain-bandwidth product and loop stability margin on the cancellation depth and bandwidth. The end result is a 26.0 dB improvement in the low-frequency band, and this improvement does not decay with frequency from 100 kHz to 10 MHz.

[0099] The second causal chain originates from the excitation mechanism of parasitic resonance. The energy of the resonance peak comes from the excitation of the parasitic loop at the instant of the transition. Two complementary transitions in the same bridge arm will excite the same loop sequentially, and the time interval between them is the existing dead time. By tuning the dead time to a value that causes the two damped ringings to superimpose in opposite phases, one excitation cancels out the other, at the cost of only changing the freewheeling time of the body diode, which is measured to be -0.42W. The interference period is the reciprocal of the dead time, which is 51.3MHz at a dead time of 19.51ns, which is 4.0 times the 3dB bandwidth of the resonance peak. Therefore, one interference zero point can cover the entire resonance peak. The key to this chain is that the time scale of the control quantity must match the frequency domain width of the resonance peak. If a time interval of three orders of magnitude larger than the conduction time is used instead, the interference period will be reduced to 238kHz, with the spectral lines alternating between rises and falls while the envelope remains unchanged.

[0100] The third causal chain originates from the interaction between the two degrees of freedom. Changing the dead time alters the complex vectors of each harmonic through the equivalent duty cycle, causing a 36.9-degree phase mismatch in the 100th harmonic, sufficient to reduce the cancellation depth from 26 dB to 4.0 dB. By applying the dead time simultaneously with amplitude scaling and phase shifting using a feedforward approach, the number of iterations required for convergence is reduced from 23 to 7. Furthermore, resource allocation for the two degrees of freedom is based on maximizing the minimum margin across the entire frequency band, yielding the counterintuitive conclusions that the optimal dead time is not at the resonant zero point and the optimal ringing amplitude ratio is not at the fully balanced point, achieving an additional 6.0 dB margin for the relatively accurate aligned solution.

[0101] The three causal chains work together to improve the minimum margin across the entire frequency band in automotive, industrial, and consumer electronics scenarios from -16.8dB, -23.0dB, and -14.9dB to +1.3dB, +3.0dB, and +11.1dB, respectively. The number of out-of-specification frequency points is reduced from 128, 75, and 3 to 0, while the conversion efficiency decreases by no more than 0.067 percentage points, and the closed-loop convergence time is no more than 9.3ms. The method does not increase the size of magnetic components or the components in the power path, and the active suppression unit can share clock and computing resources with existing digital controllers.

[0102] It should be noted that the first and second switching transistors described in this invention are not limited to synchronous buck topologies. Any topology with two complementary switching transistors in the same bridge arm and a dead time between them, including totem-pole power factor correction circuits, active clamp flyback circuits, half-bridge and full-bridge DC-DC converter circuits, three-phase inverter circuits, etc., can implement this method. For three-phase inverter circuits, the peak frequency points of the three bridge arms can be identified and the dead time can be set separately. The injection point is not limited to the positive wire at the power input terminal; any position where the driving point impedance of the injection path remains stable within the measurement window at the target frequency can be used as the injection point.

[0103] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An active suppression method for electromagnetic compatibility interference in new energy switching power supplies, characterized in that, include: The conducted interference signal is synchronously acquired using the switching clock of the switching power supply as a coherent reference to obtain the complex vector of each switching harmonic component and identify the peak frequency point formed by parasitic parameter resonance in the interference spectrum. For the selected target harmonic order, the amplitude and phase of the sinusoidal component that is equal in amplitude and opposite in phase to the target harmonic component at the injection point are determined by the complex vector of the target switching harmonic component and the complex gain of the injection path at the target harmonic frequency. The sinusoidal components are summed and then injected into the input terminal of the switching power supply through the injection path. The dead time between the first switch being turned off and the second switch being turned on in the same bridge arm is determined based on the peak frequency point, so that the damping ringing excited by the first switch being turned off and the damping ringing excited by the second switch being turned on are superimposed in opposite phase at the peak frequency point; The equivalent duty cycle change caused by the change in dead time is used to feed forward the complex vector of each switching harmonic component. Based on the reacquired complex vector, the amplitude and phase of each sinusoidal component and the dead time are iteratively updated until the interference amplitude at each frequency point is lower than the preset limit.

2. The active suppression method for electromagnetic compatibility interference in new energy switching power supplies according to claim 1, characterized in that, The method of synchronously acquiring conducted interference signals using the switching clock of the switching power supply as a coherent reference to obtain the complex vectors of each switching harmonic component includes: using the conduction edge of the switching transistor drive signal as the sampling trigger reference, sampling the conducted interference signal output by the line impedance stabilization network at a sampling rate that is an integer multiple of the switching frequency in each switching cycle to obtain a single-cycle sampling sequence; accumulating the single-cycle sampling sequence for M consecutive switching cycles point by point according to the indices within the cycle to obtain a coherent accumulation sequence, where M is 16 to 256; performing a discrete Fourier transform on the coherent accumulation sequence, and taking the complex values ​​at integer multiples of the switching frequency as the complex vectors of the corresponding switching harmonic components, where the magnitude of the complex value is the amplitude of the switching harmonic component and the argument of the complex value is the phase of the switching harmonic component.

3. The active suppression method for electromagnetic compatibility interference in new energy switching power supplies according to claim 1, characterized in that, The complex gain of the injection path at the harmonic frequency is obtained by online identification, which includes: applying a probe component with known amplitude and phase to the injection path at the harmonic frequency; acquiring the complex vector of the switching harmonic component before and after applying the probe component, and taking the difference between the two as the response change; using the ratio of the response change to the probe component as the complex gain of the injection path at the harmonic frequency; and iterating through each of the target harmonic orders to obtain the complex gain at each target harmonic frequency.

4. The active suppression method for electromagnetic compatibility interference in new energy switching power supplies according to claim 3, characterized in that, Also includes: The feedforward neural network outputs a correction amount for the complex gain. The input of the feedforward neural network is a feature vector, which includes the waveform distortion rate of the input current within the power frequency cycle, the equivalent impedance surrogate at the input terminal, the output current of the switching power supply, the case temperature of the switching transistor, and the residual amplitude at each of the target harmonic orders. A loss function is constructed using the sum of the squares of the residual amplitudes at each of the target harmonic orders. During closed-loop operation, the weights of the feedforward neural network are updated online using the backpropagation algorithm. When the change in the equivalent impedance surrogate at the input terminal relative to its value during online identification exceeds a preset threshold, the complex gain is corrected with the correction amount without re-executing the online identification.

5. The active suppression method for electromagnetic compatibility interference in new energy switching power supplies according to claim 1, characterized in that, The step of determining the dead time between the turn-off of the first switch and the turn-on of the second switch in the same bridge arm based on the peak frequency includes: taking twice the circumferential angle of the product of the peak frequency and the dead time, plus the difference in excitation phase between the turn-off of the first switch and the turn-on of the second switch on the damped ringing, which is equal to an odd multiple of the circumferential angle as an anti-phase condition; solving the anti-phase condition to obtain a set of candidate dead time values; and selecting from the set the dead time that falls within a preset dead time allowable interval and has the smallest difference from the current dead time.

6. The active suppression method for electromagnetic compatibility interference in new energy switching power supplies according to claim 5, characterized in that, Also includes: A single switching transition is divided into at least two driving segments with different gate drive currents. The ratio of the gate drive currents of each driving segment is used to adjust the amplitude ratio of the damped ringing excited by the second switch being turned on to the damped ringing excited by the first switch being turned off. The ratio of the dead time to the amplitude is used as a two-dimensional optimization variable. The goal is to maximize the minimum of the differences between the interference amplitudes at each frequency point and the preset limit. The optimal solution is searched and obtained within the preset dead time allowable interval and the preset amplitude ratio interval.

7. The active suppression method for electromagnetic compatibility interference in new energy switching power supplies according to claim 1, characterized in that, The method of feedforward correction of the complex vector of each switching harmonic component using the equivalent duty cycle change caused by the change in dead time includes: obtaining the equivalent duty cycle change by the ratio of the change in dead time to the switching period; scaling the amplitude of the complex vector of each switching harmonic component according to the ratio of the Sink function values ​​at the equivalent duty cycle before and after correction; and shifting the phase of the complex vector of each switching harmonic component according to the circumferential angle of the product of the harmonic order and the equivalent duty cycle change; and applying the feedforward correction while applying the dead time in this iteration.

8. The active suppression method for electromagnetic compatibility interference in new energy switching power supplies according to claim 1, characterized in that, The iterative update further includes: using the difference between the interference amplitude of each frequency point and the preset limit as the margin of that frequency point, and taking the smallest margin among the frequency points as the evaluation quantity; when the margin at the target harmonic number increases, the margin at other frequency points decreases, and the evaluation quantity does not increase in the current iteration, the target harmonic number whose margin is greater than the preset threshold is removed from the target harmonic number, and the harmonic number corresponding to the frequency point where the evaluation quantity is located is moved into the target harmonic number; the maximization of the evaluation quantity is used as the iteration termination criterion.

9. The active suppression method for electromagnetic compatibility interference in new energy switching power supplies according to claim 1, characterized in that, Also includes: Set an upper limit for the amplitude of the signal obtained by summing the sinusoidal components. When the amplitude of the summed signal exceeds the upper limit, the amplitude is allocated in ascending order of margin at each target harmonic order until the upper limit is reached. If the residual amplitude at each target harmonic order does not decrease in two consecutive iterations, the amplitude and phase of each sinusoidal component are reverted to the values ​​of the previous iteration. Using the on-state voltage drop of the switching transistor as a junction temperature proxy, when the junction temperature proxy exceeds a preset threshold, the update of the dead time is frozen and the current value of the dead time is maintained.

10. The active suppression method for electromagnetic compatibility interference in new energy switching power supplies according to claim 1, characterized in that, Also includes: The corresponding limit template and frequency band weights are loaded according to the application scenario of the switching power supply. The application scenarios include automotive, industrial, and consumer electronics scenarios. The preset limit is given by the loaded limit template. In the automotive scenario, the switching harmonic components falling within the radio broadcast receiving frequency band are preferentially selected as the target harmonic order, and the frequency band weight of the radio broadcast receiving frequency band is greater than the frequency band weight of other frequency bands. When the peak frequency is higher than the upper limit of the loaded limit template, the dead time is kept unchanged at the value tuned according to efficiency, and the number of target harmonic orders is increased to the upper limit. In the iterative update, the difference between the interference amplitude of each frequency point and the loaded limit template is weighted according to the loaded frequency band weight.

Images