Redundant switching power supply system for data center server power supply

Abstract

The application discloses a redundant switching power supply system for data center server power supply, and relates to the technical field of data center power supply.The redundant switching power supply system comprises a load pre-judging unit, a time delay calibration unit in each power module, a compensation controller, an impedance matching network, an oscillation suppression unit and a feedback fine adjustment unit.The load pre-judging unit acquires a pulse width modulation instruction, predicts load current variation, and generates a current compensation pre-instruction.The time delay calibration unit in each power module calibrates transmission time delay and line parasitic inductance.The compensation controller performs time base translation on the pre-instruction according to the transmission time delay, so that compensation currents of the modules are synchronized to reach a load bus.The impedance matching network adjusts output impedance according to the parasitic inductance.The oscillation suppression unit collects bus high-frequency ripples and performs filtering processing.The feedback fine adjustment unit monitors residual ripples, and when the amplitude of the residual ripples exceeds a preset threshold, sends a correction instruction to the compensation controller and the impedance matching network to guide the two to cooperatively adjust.The application improves bus voltage stability and load end power supply quality.

Description

Technical Field

[0001] This invention relates to the field of data center power supply technology, specifically to a redundant switching power supply system for powering data center servers. Background Technology

[0002] Redundant switching power supply technology is a core component for ensuring high availability of data center servers. Current mainstream redundant power supply solutions typically employ an N+M parallel current sharing architecture, which can be implemented in two main ways: First, there's the active current sharing method based on analog control. This method samples the output current of each power module in real time and uses operational amplifiers to adjust the feedback loop to achieve load balancing. This method offers rapid response, but due to the discreteness of analog devices, it's prone to circulating current oscillations when multiple modules are connected in parallel, and it struggles to adapt to the millisecond-level fluctuations in processor load. Second, there's the current sharing strategy based on digital management. This method coordinates the output of each module using an inter-integrated circuit bus or power management bus. While this improves monitorability and flexibility, communication delays and sampling quantization errors often lead to excessive voltage drops or severe current contention between modules during dynamic load switching. However, existing technologies generally focus on improving current sharing accuracy and isolating short-circuit faults under steady-state conditions, neglecting the deep-seated inherent problem of dynamic response phase mismatch caused by physical layout. In practical engineering, the line impedance and parasitic inductance of power modules in different slots to the load bus vary significantly. This causes the compensation current from the far-end module to lag behind that from the near-end module when it reaches the load bus. While existing patents address OR gate protection or hot-swap suppression, they do not disclose how to solve the dynamic compensation competition and high-frequency parasitic oscillation problems caused by bus length delays. Furthermore, they do not address the ripple interference caused by this oscillation on the load bus voltage, or its impact on the input of the subsequent voltage regulation module. Therefore, how to fundamentally address the transient response synchronization issues caused by physical layout differences among multiple modules under large dynamic loads, while ensuring steady-state current sharing accuracy, and eliminate the resulting false protection and output instability, remains a critical technological blind spot that urgently needs to be overcome. Summary of the Invention

[0003] The purpose of this invention is to provide a redundant switching power supply system for powering data center servers, in order to solve the problems mentioned in the background art.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a redundant switching power supply system for powering data center servers, comprising multiple power modules, each power module being connected in parallel to the load bus, and further comprising: The load prediction unit is used to obtain the pulse width modulation command of the voltage regulation module, predict the change amount and rate of change of the load current according to the command, and generate the current compensation pre-compensation command for each power supply module. A time delay calibration unit is installed in each power module to calibrate the transmission time delay between the power module and the load bus. The compensation controller is installed in each power module. It receives the current compensation pre-compensation command and performs time base shifting on the current compensation pre-compensation command according to the transmission delay of the power module, so that the compensation current of each power module arrives at the load bus at the same time. An impedance matching network is set between the output terminal of each power module and the load bus. Based on the line impedance parameters between the output terminal of the power module and the load bus, the output impedance of the power module is adjusted. An oscillation suppression unit is connected between the near and far ends of the load bus to collect high-frequency ripple signals on the load bus and to filter the high-frequency ripple signals. The feedback fine-tuning unit is connected to the oscillation suppression unit and is used to receive the residual ripple signal after the oscillation suppression unit has filtered it. When the amplitude of the residual ripple signal exceeds a preset threshold, it sends a correction command to the compensation controller and the impedance matching network. Each power module outputs compensation current to the load bus through the impedance matching network according to the current compensation pre-instruction after time base shift. The compensation current of each power module propagates through its respective transmission path and arrives at the load bus simultaneously.

[0005] Preferably, the load prediction unit is connected to the pulse width modulation controller of the voltage regulation module via a communication bus, and the load prediction unit reads the duty cycle instruction from the pulse width modulation controller.

[0006] Preferably, the delay calibration unit sends an identification signal to the load bus when the system is powered on, receives the echo of the identification signal, and calibrates the transmission delay from the power module to the load bus based on the time difference between the sending and receiving times.

[0007] Preferably, the time delay calibration unit is also used to calibrate the parasitic inductance of the line between the output terminal of the power module and the load bus, and output the parasitic inductance as the line impedance parameter to the impedance matching network.

[0008] Preferably, the impedance matching network includes a variable resistor array and a switched capacitor array. The resistance value of the variable resistor array and the capacitance value of the switched capacitor array are adjusted according to the parasitic inductance so that the output impedance of the power module approaches a preset target impedance value.

[0009] Preferably, the method by which the compensation controller performs time base shift on the current compensation pre-command is as follows: obtain the maximum value of the transmission delay in all power modules, calculate the difference between the maximum value and the transmission delay of the power module, and use the difference as the time base shift amount of the power module.

[0010] Preferably, the oscillation suppression unit includes an adjustable filter, the filtering parameters of which are set according to the transmission delay and line impedance parameters of each power module, and are used to filter the high-frequency ripple signal on the load bus.

[0011] Preferably, when the amplitude of the residual ripple signal exceeds a preset threshold, the feedback fine-tuning unit sends a correction command to each compensation controller, and each compensation controller readjusts its own time base shift according to the correction command.

[0012] Preferably, the feedback fine-tuning unit also sends correction commands to each impedance matching network, and each impedance matching network readjusts its output impedance according to the correction commands.

[0013] Preferably, the load bus is a DC bus with a nominal voltage of 48 volts.

[0014] This invention provides a redundant switching power supply system for powering data center servers. It offers the following advantages: By setting up a time delay calibration unit and a compensation controller in each module, each module can measure its own transmission delay to the load bus upon power-up. The compensation controller calculates the time base shift for each module individually based on the maximum time delay value among all modules. The far-end modules send out the compensation current in advance, while the near-end modules send out the compensation current later. In this way, although the sending times of each module are different, the compensation current can reach the load end simultaneously after propagating along the bus, eliminating the dynamic response lag problem caused by differences in physical layout. At the same time, the time delay calibration unit can also measure the parasitic inductance of each module's line. The impedance matching network adjusts the variable resistor and switched capacitor according to the inductance value to make the output impedance of each module tend to be consistent. With the combined effect of time synchronization and impedance consistency, each module can work together to exert force when the dynamic load changes, and there will be no situation where the near-end module is overloaded and the far-end module cannot compensate. The bus voltage can recover to stability as soon as possible when the load changes.

[0015] The oscillation suppression unit is connected between the near and far ends of the busbar, collecting high-frequency ripple in the differential voltage and filtering it according to the time delay and impedance parameters of each module. The feedback fine-tuning unit monitors the residual ripple after filtering. When the ripple amplitude exceeds the preset threshold, it simultaneously sends correction commands to the compensation controller and impedance matching network, guiding them to adjust in tandem. The compensation controller fine-tunes the time base shift, and the impedance matching network fine-tunes the output impedance. The two work together to converge repeatedly, and the high-frequency oscillation of the busbar is gradually reduced. After the busbar voltage stabilizes, the ripple at the input of the voltage regulation module will also decrease, and the operating conditions of the voltage regulation module will be improved. The entire system does not rely on conventional methods such as increasing the switching frequency or speeding up the loop response. Instead, it starts from the source of synchronous compensation and impedance matching, and has good adaptability to different cabinet layouts and module configurations. Attached Figure Description

[0016] Figure 1 The feedback fine-tuning and collaborative control logic diagram of the redundant switching power supply system for powering data center servers in this invention is shown below. Figure 2 This is a flowchart illustrating the process of setting the center frequency of the band-stop filter according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 and Figure 2This invention provides a redundant switching power supply system for powering data center servers. In this system, the voltage regulation module is a DC-DC converter connected between the load bus and the server load. Internally, it contains a pulse width modulation (PWM) controller that outputs a corresponding PWM duty cycle command based on load requirements. During operation, the load prediction unit communicates with the voltage regulation module. The voltage regulation module contains a PWM controller that outputs a duty cycle command, reflecting the current requirement of the processor in the next moment. Upon receiving this command, the load prediction unit performs internal calculations to anticipate changes in the load current. The load prediction unit maintains a mapping table recording the output current values ​​corresponding to different PWM duty cycle values. The load prediction unit compares the current duty cycle value with the previous duty cycle value to calculate the amount and direction of the duty cycle change. Subsequently, the load prediction unit uses the current values ​​corresponding to the two duty cycles adjacent to the current duty cycle in the mapping table to calculate the load current change corresponding to the duty cycle change. Simultaneously, it divides the duty cycle change by the time interval between the two read operations to obtain the load current change rate. After completing the above calculations, the load prediction unit determines the current change trend for this compensation based on the load current change and its rate of change. The load prediction unit reads the current measured bus voltage value and, based on the direction of the deviation between this measured value and the nominal bus voltage, determines the polarity of the current compensation, i.e., whether to increase or decrease the output current. Based on the load current change and its rate of change, combined with the current measured bus voltage value, and following the principle of power balance (i.e., the bus output power and load power consumption are balanced), the load prediction unit calculates the total current required to maintain bus voltage stability, based on the current measured bus voltage value and the load current change, and thus obtains the total current required to be supplemented by each power module. After obtaining the total current, the load prediction unit distributes the total current evenly among the power modules currently online in the system, based on the number of modules, forming a current compensation pre-instruction for each module. This pre-instruction contains three pieces of information: the target amplitude of the compensation current, the rate of change of the compensation current, and the time when the pre-instruction is issued. The load prediction unit simultaneously sends this pre-instruction to the compensation controllers of all online power modules via an internal communication link. Based on this prediction result, the load prediction unit generates a current compensation pre-instruction for each power module, informing each module how much compensation current to output next.

[0019] Upon initial system power-up, the time delay calibration unit within each power module begins operation. This unit sends an identification signal to the load bus, which propagates along the bus and reflects back at its far end. The identification signal uses a square wave pulse with a rise time controlled in the nanosecond range, a pulse width in the microsecond range, and an amplitude set to half the nominal bus voltage. This pulse type is chosen to ensure a steep rise time for the echo signal, facilitating subsequent timing. The time delay calibration unit contains a high-speed comparator with its reference level set at the midpoint of the pulse amplitude. When the echo signal enters the comparator, its output flips when the signal amplitude crosses the reference level. The time delay calibration unit records this flip as the echo arrival time. Using a comparator to capture the echo arrival time eliminates timing measurement errors caused by signal amplitude fluctuations.

[0020] The delay calibration unit records the pulse emission time and the echo arrival time. Using the time difference between these two times, the transmission delay between the power module and the load bus can be calculated. Different power modules are installed in different locations in the rack; some are closer to the load, and some are farther away. Therefore, the transmission delay measured for each module is different, with the delay value of the far-end module being larger than that of the near-end module.

[0021] This transmission delay value is sent to the compensation controller. After receiving the current compensation pre-command from the load prediction unit, the compensation controller does not execute it immediately, but first performs a time base shift. Specifically, the compensation controller first checks which power module has the largest transmission delay, then subtracts its own module's transmission delay from this maximum value to calculate a difference. This difference is the amount of time that module needs to send the command in advance. Remote modules, due to their larger transmission delays, have smaller calculated differences and almost no need for advance sending; near-end modules, with their smaller transmission delays, have larger calculated differences and need to wait longer before sending the command. After this processing, although each module sends the compensation command at a different time, these compensation currents, after actually propagating along the bus, will arrive at the load end at the same moment.

[0022] Simultaneously with system power-on, the time delay calibration unit performs another function: calibrating the parasitic inductance between the power module's output and the load bus. This parasitic inductance value is then sent to the impedance matching network. The impedance matching network, located between the power module's output and the load bus, consists of variable resistors and switched capacitors. Upon receiving the parasitic inductance value, the impedance matching network adjusts the resistance of the variable resistors and the capacitance of the switched capacitors accordingly, bringing the power module's output impedance closer to a preset target impedance value. This target impedance value is a fixed impedance value pre-set during system design. The selection principle is to match the power module's output impedance with the characteristic impedance of the load bus, suppressing reflections of the compensation current propagating on the bus. All power modules share the same target impedance value, thus ensuring that the equivalent impedance of each module looking from its output to the bus is consistent.

[0023] The adjustment method is as follows: The impedance matching network stores a mapping table internally. The mapping table is generated as follows: During the system design phase, the actual impedance characteristics of bus cables of different lengths and specifications under various parasitic inductance values ​​are measured under laboratory conditions. The measured impedance characteristics are compared with the target impedance value to determine the resistance and capacitance combination that makes the output impedance closest to the target impedance value. This combination is then associated with and stored with the corresponding parasitic inductance value range.

[0024] Each row of the mapping table records a range of parasitic inductance values ​​and the corresponding resistance and capacitance adjustment levels. Upon receiving a parasitic inductance value, the impedance matching network first searches the mapping table for the range to which the value belongs, then retrieves the corresponding resistance and capacitance adjustment levels. The variable resistor array has multiple levels, each corresponding to a resistance value; level switching is achieved by turning internal switches on or off. Similarly, the switched capacitor array has multiple levels, each corresponding to a capacitance value; level switching is also achieved by turning internal switches on or off. The impedance matching network simultaneously adjusts the variable resistor array and the switched capacitor array according to the retrieved levels, making their combined impedance approach the preset target impedance value. This mapping table is written into the impedance matching network's memory at the factory. During system operation, the impedance matching network actively adjusts its output impedance using this mapping table based on the latest parasitic inductance value obtained from the time delay calibration unit. Different modules have different parasitic inductances, so their adjustment ranges also differ. The ultimate goal is to ensure that the equivalent impedance of all modules looking from the output end to the bus is essentially the same.

[0025] After time base shifting and impedance matching, each power module outputs compensation current according to the shifted time point, via the pre-adjusted impedance matching network, ensuring that the compensation current from each module arrives at the load bus simultaneously. Because the compensation currents from each module arrive synchronously and their output impedances are now nearly identical, there will be no situation where some modules output excessive power while others output insufficient power when the compensation currents converge on the bus. Once the compensation current reaches the load, it effectively compensates for changes in the server load's current demand, thus maintaining the bus voltage within the specified range.

[0026] The oscillation suppression unit is connected between the near and far ends of the load bus, and it collects high-frequency ripple signals appearing on the bus in real time. After collecting the ripple signal, the oscillation suppression unit filters the collected ripple based on the transmission delay and line impedance parameters of each power module, filtering out the high-frequency parasitic oscillation components. If there is still residual ripple signal after filtering, and the amplitude of the residual ripple exceeds a preset threshold, the feedback fine-tuning unit will intervene. The feedback fine-tuning unit sends a correction command to each compensation controller, which readjusts the time base shift upon receiving the command. At the same time, the feedback fine-tuning unit also sends a correction command to each impedance matching network, which readjusts the output impedance. Through this repeated fine-tuning, the oscillation is gradually reduced and will not continue to accumulate on the bus.

[0027] In actual rack deployments, the load prediction unit can be a standalone control board, plugged into the server's management slot, or integrated within the power management module. Regardless of the physical form, the communication bus connection between it and the pulse width modulation controller is the same, achieved through a standard bus interface and cable. The type of communication bus is already well-established in engineering, requiring only that the data transmission rate be sufficient to keep pace with the switching frequency of the pulse width modulation controller.

[0028] Furthermore, other devices that need to read pulse width modulation information can be connected to the communication bus without affecting the normal reading operation of the load prediction unit. The load prediction unit has its own independent address on the bus. The duty cycle command data issued by the pulse width modulation controller is broadcast to the bus, and the load prediction unit retrieves the required data fields according to its own address, without conflicting with the data of other devices.

[0029] Each power module contains a time delay calibration unit, which is responsible for measuring the transmission path length between the module and the load bus. When the system is first powered on, before all power modules have started supplying power to the load, the time delay calibration unit will actively send an identification signal to the load bus.

[0030] The identification signal originates from the output of the power module and travels along the load bus to the far end. The far end of the bus is open, and the signal reflects back after reaching its destination, forming an echo signal. The time delay calibration unit continuously monitors the bus; it records the moment it sends out the identification signal, and records the arrival time of the echo signal when it reappears on the bus. There is a time difference between these two moments; this time difference is the time required for the signal to travel one round trip on the bus.

[0031] Because the speed of signal propagation within the busbar is constant, the total distance the signal travels can be calculated using the round-trip time difference. Half of this total distance is the distance between the power module and the far end of the busbar. However, what we are truly concerned with here is not the physical distance itself, but rather the time it takes for the signal to travel from the module to the load end; this time is the transmission delay. In practice, the delay calibration unit does not need to convert the time difference into distance; instead, it directly uses the time difference value as the transmission delay.

[0032] Different power modules are installed in different positions within the rack; some are close to the load, while others are far from it. Modules closer to the load have shorter signal travel times and thus lower measured transmission delays; modules farther from the load have longer signal travel times and thus higher measured transmission delays. Each module is measured independently without interference.

[0033] To ensure the accuracy of the measured time delay value, the identification signal is selected with a waveform featuring a steep rising edge, making it easy to accurately capture the arrival time of the echo. The time delay calibration unit contains an internal clock counter with a sufficiently high counting frequency to distinguish subtle time differences. The measured transmission delay value is stored in the module's local register for later use by the compensation controller. After a measurement is completed upon power-on, these values ​​are retained until the system is re-powered or a recalibration command is received.

[0034] In addition to measuring transmission delay, the delay calibration unit also performs another measurement during the same power-on process: calibrating the parasitic inductance of the line between the power module and the load bus. Parasitic inductance is an objective physical quantity, mainly derived from the distributed inductance of the connecting cable from the power module output to the bus and the bus section itself.

[0035] When the time delay calibration unit sends the identification signal to the bus, the signal is affected by the line inductance during propagation. Inductance impedes rapidly changing current, which manifests as a slower rising edge in the signal waveform. The time delay calibration unit contains a waveform capture circuit that records both the sent identification signal and the returned echo signal. By comparing the differences in the rising edge shape between the transmitted and echo signals, and combining this with the previously measured transmission delay value, the delay calibration unit can calculate the parasitic inductance of a bus path. The calculation method is as follows: The delay calibration unit measures the time it takes for the rising edge of the echo signal to rise from the low end to the high end of the amplitude; this time is called the rise time. The rise time of the transmitted signal is a known fixed value. There is a difference between the rise time of the echo signal and the rise time of the transmitted signal, which is mainly caused by the parasitic inductance on the bus path. The delay calibration unit divides this difference by the transmission delay value to obtain a ratio, and then multiplies this ratio by a preset conversion factor to obtain the estimated value of the parasitic inductance of the bus path. The delay calibration unit internally stores the conversion factors corresponding to different cable types. When the system is powered on, the corresponding coefficient is selected for calculation based on the actual cable type used.

[0036] This calculation process is completed automatically within the time delay calibration unit, requiring no external intervention. The calibrated parasitic inductance value is converted into a line impedance parameter, which serves as the basis for the subsequent impedance matching network to adjust its output impedance. The time delay calibration unit transmits this impedance parameter directly to the impedance matching network within the same power module via internal signal lines.

[0037] The parasitic inductance measured for different power modules varies. Within a server rack, modules closer to the busbar have shorter cables and therefore lower parasitic inductance; modules farther from the busbar have longer cables, and combined with the accumulated inductance of the busbar itself, their parasitic inductance is higher. These differences are real and cannot be eliminated by simply shortening the cables, as the physical layout of the server rack is fixed at the factory.

[0038] After the time delay calibration unit completes the parasitic inductance calibration, it also stores this value in a local register. If the system temperature changes during operation, the physical characteristics of the cable will change slightly, and the parasitic inductance will also change accordingly. At this time, an external command can be used to trigger the time delay calibration unit to recalibrate, update the stored value, and ensure that the impedance matching network always has accurate parameters as a basis for adjustment.

[0039] An impedance matching network is installed between the output of the power module and the load bus, connected in series in the path of the incoming current. This matching network consists of two parts: a variable resistor array and a switched capacitor array. The variable resistor array consists of multiple resistors and switching transistors. By controlling the switching transistors, different resistors can be connected to or disconnected from the circuit, thereby changing the total resistance value. The switched capacitor array consists of multiple capacitors and switching transistors. By controlling the switching transistors, the total capacitance value connected to the circuit can be changed.

[0040] After receiving the current compensation pre-compensation command from the load prediction unit, the compensation controller does not immediately adjust its output according to the command. The first thing the compensation controller does is acquire the transmission delay values ​​of all power modules. These values ​​have already been measured by the delay calibration units of each module when the system is powered on and are stored in their respective local registers. The compensation controller reads the transmission delay values ​​of all other modules through the internal communication links between the modules.

[0041] After obtaining the transmission delay of all modules, the compensation controller will find the largest value among these values. This maximum value represents the module that is farthest from the load, and the module from which the signal takes the longest to reach the load. The compensation controller finds this maximum value and then subtracts the transmission delay of its own module from this maximum value. The difference is the time base shift of that module.

[0042] What does this time base shift mean? If a module is very close to the load, its transmission delay is very small, then the difference between the maximum value and its delay will be relatively large. After receiving the pre-command, this module needs to wait a relatively long time before sending the compensation current, because the current takes less time to travel from the near end to the load; sending it earlier would result in the current arriving earlier and not matching the current of other modules. Conversely, if a module is very far from the load, its transmission delay is inherently large, then the difference between the maximum value and its delay will be very small. This module hardly needs to wait and sends it out quickly after receiving the pre-command, because the current takes a long time to travel from the far end to the load. In this way, each module schedules its transmission time according to its own unique waiting time.

[0043] After calculating the time base shift, the compensation controller will hold the current compensation pre-command for a period of time, the length of which is the difference calculated earlier. Once the holding time is up, the compensation controller will release the current compensation pre-command, allowing the subsequent execution circuits to start operating. In this way, the remote module sends the command first, followed by the near-end module. Although the timing of the commands is different, the compensation current will reach the load location at the same time after propagating along the bus.

[0044] During system operation, if a power module is disconnected due to a fault or a new module is inserted, the latency distribution of the entire system will change. The compensation controller continuously monitors the module status during system operation. Each power module has an internal status register that records whether the module is working properly. The compensation controller reads the values ​​of all module status registers at fixed intervals via an internal communication link. When the compensation controller detects that a module's status changes from normal to abnormal, it determines that the module is offline; when it detects that a module's status changes from abnormal to normal, it determines that a new module has come online. In the next cycle after detecting the above changes, the compensation controller re-acquires the transmission latency values ​​of all currently online modules, finds the maximum value, and recalculates its own time base shift. From detecting the module change to completing the time base shift update, the entire process is completed within several switching cycles. This dynamic update mechanism ensures that even if the system configuration changes, the compensation current of each module can still arrive synchronously.

[0045] The oscillation suppression unit is structurally connected between the near and far ends of the load bus. Its input signal is taken from two voltage sampling points at different locations on the bus; the near-end sampling point is close to the output junction of the power module, and the far-end sampling point is close to the load connection of the server. The oscillation suppression unit simultaneously collects the voltage signals at these two locations, then performs differential processing to extract the high-frequency fluctuation components present on the bus from the differential result.

[0046] After all power modules synchronously output compensation current, some high-frequency fluctuations may appear on the bus, which are superimposed on the DC bus voltage. The causes of these fluctuations are complex and related to the switching actions of each module, the parasitic parameters of the bus, and the dynamic changes of the load. The differential signal acquired by the oscillation suppression unit contains this fluctuation information. The oscillation suppression unit has an internal filter circuit that can separate the high-frequency components from the overall signal.

[0047] The oscillation suppression unit's filtering process is not performed using a fixed filter. The filter parameters are set with reference to the transmission delay values ​​and line impedance parameters of each power module. Different modules have different delays and impedances, resulting in varying frequency distributions of oscillations on the bus. The oscillation suppression unit obtains the current settings for the transmission delay and impedance matching network from the compensation controllers of each module, and then adjusts the filter's operating parameters accordingly. The filter employs a band-stop structure, which allows low-frequency components to pass while blocking high-frequency components near specific frequencies. The center frequency of the band-stop filter determines which frequency components it blocks. After obtaining the transmission delay values ​​and current settings for the impedance matching network from each module, the oscillation suppression unit first calculates the average transmission delay of all currently online modules and the average output impedance of the impedance matching network of all currently online modules. Then, based on the calculated average transmission delay and average impedance values, the oscillation suppression unit searches for the resonant fundamental frequency position corresponding to the current average transmission delay and average impedance values ​​in its internally stored mapping table. The mapping table is generated as follows: During the system design phase, buses of different lengths and impedance characteristics are tested under laboratory conditions. The actual resonant frequencies corresponding to each set of transmission delay and impedance parameters are recorded, and the resonant frequencies are associated and stored with the corresponding delay and impedance parameters. After the oscillation suppression unit looks up the resonant fundamental frequency position in the current state, it sets the center frequency of the band-stop filter to the value of that resonant frequency. After processing by the band-stop filter, the high-frequency fluctuation components near the center frequency are significantly weakened, while other frequency components are basically unaffected. If the oscillation suppression unit finds that there is still obvious residual fluctuation after filtering, it indicates that there is a deviation between the actual oscillation frequency and the preset center frequency. The oscillation suppression unit will fine-tune the center frequency in a step-by-step manner, adjusting by a fixed frequency increment each time, gradually approaching the actual oscillation frequency until the residual fluctuation amplitude is significantly reduced. During the fine-tuning process, the oscillation suppression unit continuously monitors the direction of change of the residual fluctuation amplitude. If the residual fluctuation amplitude increases after a certain adjustment, it will adjust in the opposite direction to ensure that the convergence direction is correct.

[0048] After filtering, the oscillation suppression unit outputs a more stable signal, free from fluctuations. This output signal can be used to assess the current stability of the bus. Simultaneously, the oscillation suppression unit continuously monitors the difference between the input and output signals. If the difference exceeds a preset deviation range, it indicates that the existing filtering parameters may be inappropriate, and the oscillation suppression unit will readjust the filter settings.

[0049] The feedback fine-tuning unit continuously receives the signal output from the oscillation suppression unit. After completing the filtering process, the oscillation suppression unit sends the residual ripple signal separated during the processing to the feedback fine-tuning unit. This residual ripple signal reflects the high-frequency fluctuations that remain on the bus and have not been completely eliminated by the filter.

[0050] The feedback fine-tuning unit incorporates a comparison mechanism that measures the amplitude of the residual ripple signal in real time. The system is pre-defined with a threshold value representing the upper limit of permissible high-frequency fluctuations on the bus. The feedback fine-tuning unit compares the measured residual ripple amplitude with this threshold. If the residual ripple amplitude is within the threshold, it indicates that the high-frequency fluctuations on the bus are within acceptable limits, and the feedback fine-tuning unit performs no further action, continuing its monitoring.

[0051] If the residual ripple amplitude exceeds the threshold, it indicates that there is still significant oscillation on the bus. At this point, the feedback fine-tuning unit will intervene. The feedback fine-tuning unit sends correction commands to the compensation controllers of each power module through the inter-module communication link. The correction command includes two pieces of information: adjustment direction and adjustment step size. The direction information includes two values: increasing or decreasing the translation amount. The step size information is a fixed time increment value, which is preset in the feedback fine-tuning unit during system design. After receiving the correction command, each compensation controller decides whether to increase or decrease its own time base translation amount according to the direction information, and determines the magnitude of each adjustment according to the step size information. The correction command will be sent continuously multiple times until the residual ripple amplitude falls back below the threshold. If the residual ripple amplitude does not change significantly after several consecutive adjustments, the feedback fine-tuning unit will determine that the current adjustment direction may be incorrect, and then switch the adjustment direction and resend the correction command.

[0052] After receiving the correction command, the compensation controller of each power module will readjust its time base shift according to the command requirements. The adjustment range is determined by the compensation controller internally based on the current actual state. The feedback fine-tuning unit only provides the adjustment direction, and the specific adjustment amount is calculated by each compensation controller itself. The reason for this is that the transmission delay of each module is different, and the adjustment range required for the same adjustment direction is also different on different modules. It is more accurate for each module to calculate it itself.

[0053] After the compensation controller completes the readjustment of the time base shift, the subsequent compensation current will be issued according to the new shift. The feedback fine-tuning unit will continue to monitor the residual ripple after the adjustment. If the ripple amplitude is still out of range, the feedback fine-tuning unit will send a correction command again to guide the compensation controller to make further fine-tuning. This process is repeated until the amplitude of the residual ripple falls back and stabilizes below the threshold.

[0054] After determining that the residual ripple amplitude exceeds the preset threshold, the feedback fine-tuning unit performs another action: simultaneously sending correction commands to each compensation controller and the impedance matching network within each power module. The correction commands sent to the impedance matching networks include the impedance adjustment direction, which can be either increasing or decreasing impedance. Upon receiving this correction command, each impedance matching network moves the current position of the variable resistor array one position in the target direction, and simultaneously moves the current position of the switched capacitor array one position in the corresponding direction. Each time the impedance matching network receives a correction command... The system shifts gears, and after each shift, it feeds back the current gear status to the feedback fine-tuning unit. The feedback fine-tuning unit determines whether the gear adjustment is effective based on the change in the residual ripple amplitude. If the amplitude is decreasing, the current adjustment direction is maintained; if the amplitude is increasing or remains unchanged, the adjustment direction is switched in the next instruction. The gear adjustment of the impedance matching network and the time base shift adjustment of the compensation controller are performed synchronously. Both receive correction instructions independently and do not affect each other. This correction instruction and the instruction sent to the compensation controller are parallel, and the two instructions are delivered to their respective targets within the same time window.

[0055] After receiving the correction command, each impedance matching network will readjust its output impedance according to the command requirements. The adjustment method is the same as the adjustment method during power-on, which is still to adjust the resistance value of the variable resistor array and the capacitance value of the switched capacitor array; the difference is that this adjustment is carried out under system load, and the impedance matching network needs to ensure that the output is not interrupted during the adjustment process.

[0056] After receiving a correction command, the impedance matching network combines the currently stored parasitic inductance values ​​with the adjustment direction given in this correction command to recalculate a new set of resistance and capacitance values. Once the calculation is complete, the impedance matching network gradually changes the resistance and capacitance connected to the circuit, rather than switching them all at once. This is done to avoid instantaneous disturbances to the bus voltage caused by sudden impedance changes. The entire switching process is completed gradually within a short time window, with each step changing only a small portion of the parameters, ensuring that the bus voltage remains within the normal range.

[0057] After sending a correction command, the feedback fine-tuning unit continues to monitor changes in residual ripple. If the residual ripple amplitude decreases after the impedance matching network adjustment but still exceeds the range, the feedback fine-tuning unit will send another correction command. This correction command includes adjustments to both the compensation controller and further adjustments to the impedance matching network. These two adjustments work together, repeatedly fine-tuning until the high-frequency fluctuations of the bus converge to an acceptable level. The adjustment of the impedance matching network and the time base shift adjustment of the compensation controller are performed simultaneously, working together to suppress bus fluctuations.

[0058] This redundant switching power supply system has a load bus, which is DC. All the power modules in the server rack connect their outputs to this bus, and the load draws power from the end of the bus. In the data center rack power supply architecture, after the AC power comes in, it is first rectified and converted into DC power before being distributed to various devices. This DC bus is used for this distribution.

[0059] The nominal voltage of this DC bus is 48 volts, which is a common practice in the data center industry. This voltage value is within the safe operating range and can transmit the same power under low current conditions, which helps to reduce line losses during transmission. The design of the entire redundant switching power supply system is centered around this 48-volt DC bus.

[0060] The output characteristics of each power module are set according to a 48V bus, and the internal switching transistors, transformers, filter capacitors, and other components are all selected based on this voltage. When the time delay calibration unit sends the identification signal, the amplitude of the identification signal is also within the range of the bus voltage and will not exceed the voltage withstand limit of the components. The variable resistors and switched capacitors in the impedance matching network are also designed with voltage withstand and power consumption based on 48V operating conditions.

[0061] Although the bus voltage may fluctuate slightly due to load changes during actual operation, the overall system control objective is to maintain the bus voltage near its nominal value. When calculating the current compensation, the load prediction unit uses 48 volts as a reference to estimate the required current to stabilize the voltage. The compensation controller, after time-base shifting, synchronizes the compensation current output by each module, ultimately acting on the bus to offset voltage deviations caused by load changes, allowing the bus voltage to return to around 48 volts. The entire closed-loop regulation of the system revolves around this voltage target.

[0062] The oscillation suppression unit is connected between the near and far ends of the bus. It also collects the high-frequency fluctuation components on this 48V DC bus. The stable signal output after filtering also reflects the actual state of this bus. The correction action made by the feedback fine-tuning unit based on the residual ripple situation ultimately serves the stability of this 48V DC bus.

[0063] During power-on initialization, the identification signal sent by the delay calibration unit is staggered with the normal power supply output. The power module only starts normal voltage and current output after the identification signal is sent and the echo is received, thus the identification process does not cause any impact on the load. During normal system operation, the delay calibration unit is in standby mode and does not actively send identification signals. It only restarts measurement when it receives an external recalibration command, thus not interfering with normal power supply output. The impedance matching network uses a step-by-step adjustment method, changing only one level at a time, and the impedance change between adjacent levels is controlled within the allowable range of bus voltage fluctuations. Therefore, the switching process does not cause significant jumps in bus voltage. The correction command of the feedback fine-tuning unit gradually converges through multiple iterations. Each adjustment is a preset fixed step size, and the adjustment direction is switched in real time according to the trend of residual ripple changes. Therefore, there will be no adjustment overshoot or oscillation divergence. All of the above ensures that the system can operate stably and reliably in a real data center environment.

[0064] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A redundant switching power supply system for powering data center servers, comprising multiple power modules connected in parallel to a load bus, characterized in that, Also includes: The load prediction unit is used to obtain the pulse width modulation command of the voltage regulation module, predict the change amount and rate of change of the load current according to the command, and generate the current compensation pre-compensation command for each power supply module. A time delay calibration unit is installed in each power module to calibrate the transmission time delay between the power module and the load bus. The compensation controller is installed in each power module. It receives the current compensation pre-compensation command and performs time base shifting on the current compensation pre-compensation command according to the transmission delay of the power module, so that the compensation current of each power module arrives at the load bus at the same time. An impedance matching network is set between the output terminal of each power module and the load bus. Based on the line impedance parameters between the output terminal of the power module and the load bus, the output impedance of the power module is adjusted. An oscillation suppression unit is connected between the near and far ends of the load bus to collect high-frequency ripple signals on the load bus and to filter the high-frequency ripple signals. The feedback fine-tuning unit is connected to the oscillation suppression unit and is used to receive the residual ripple signal after the oscillation suppression unit has filtered it. When the amplitude of the residual ripple signal exceeds a preset threshold, it sends a correction command to the compensation controller and the impedance matching network. Each power module outputs compensation current to the load bus through the impedance matching network according to the current compensation pre-instruction after time base shift. The compensation current of each power module propagates through its respective transmission path and arrives at the load bus simultaneously.

2. The redundant switching power supply system for powering data center servers according to claim 1, characterized in that: The load prediction unit is connected to the pulse width modulation controller of the voltage regulation module via a communication bus, and the load prediction unit reads the duty cycle instruction from the pulse width modulation controller.

3. The redundant switching power supply system for powering data center servers according to claim 2, characterized in that: The delay calibration unit sends an identification signal to the load bus when the system is powered on, receives the echo of the identification signal, and calibrates the transmission delay from the power module to the load bus based on the time difference between the sending and receiving times.

4. The redundant switching power supply system for powering data center servers according to claim 3, characterized in that: The time delay calibration unit is also used to calibrate the parasitic inductance of the line between the output terminal of the power module and the load bus, and output the parasitic inductance as the line impedance parameter to the impedance matching network.

5. The redundant switching power supply system for powering data center servers according to claim 4, characterized in that: The impedance matching network includes a variable resistor array and a switched capacitor array. The resistance value of the variable resistor array and the capacitance value of the switched capacitor array are adjusted according to the parasitic inductance so that the output impedance of the power module approaches the preset target impedance value.

6. The redundant switching power supply system for powering data center servers according to claim 5, characterized in that: The method for the compensation controller to perform time base shift on the current compensation pre-command is as follows: obtain the maximum value of the transmission delay in all power modules, calculate the difference between the maximum value and the transmission delay of the power module, and use the difference as the time base shift amount of the power module.

7. The redundant switching power supply system for powering data center servers according to claim 1, characterized in that: The oscillation suppression unit includes an adjustable filter whose filtering parameters are set according to the transmission delay and line impedance parameters of each power module, and is used to filter the high-frequency ripple signal on the load bus.

8. The redundant switching power supply system for powering data center servers according to claim 7, characterized in that: When the amplitude of the residual ripple signal exceeds a preset threshold, the feedback fine-tuning unit sends a correction command to each compensation controller, and each compensation controller readjusts its own time base shift according to the correction command.

9. The redundant switching power supply system for powering data center servers according to claim 8, characterized in that: The feedback fine-tuning unit also sends correction commands to each impedance matching network, and each impedance matching network readjusts its output impedance according to the correction commands.

10. The redundant switching power supply system for powering data center servers according to claim 1, characterized in that: The load bus is a DC bus with a nominal voltage of 48 volts.

Images