A transducer driving device and method with neural network predistortion

Abstract

This application discloses a transducer driving device and method with neural network predistortion. The device includes a main control unit, a full-bridge power stage, a demodulator, and an isolated sampling circuit. The method includes: a processing system running a neural network predistortion algorithm, receiving a reference input signal, a feedback actual output voltage signal, and real-time temperature data, and generating a predistortion compensation signal; a programmable logic unit generating a driving signal based on the compensation signal; the full-bridge power stage amplifying the power and driving the transducer load through the demodulator; and the isolated sampling circuit acquiring the output voltage and feeding it back to the processing system. This application can also employ a gated cyclic unit architecture with residual connections and introduce pressure and real-time impedance data as multi-dimensional feature inputs. By adjusting the dead time of the driving signal through a dynamic dead-time compensation engine, the fidelity of the driving signal, system efficiency, and operational reliability can be improved.

Description

Technical Field

[0001] This invention relates to the fields of power electronics and intelligent control, and more specifically, to a transducer driving device and method with neural network predistortion for driving transducers, particularly suitable for driving piezoelectric ceramic transducers. Background Technology

[0002] Transducers, especially piezoelectric ceramic transducers used in underwater sonar systems, are core components for underwater detection, communication, and imaging. Their performance, such as detection range and signal fidelity, is directly constrained by the power amplifier driving them. To achieve long-distance signal transmission, transducers typically require high-power sine wave or pulse signals of hundreds of watts or more to drive them. Simultaneously, sonar systems often require continuous operation for extended periods, placing stringent demands on the power amplifier's energy conversion efficiency. Inefficient power amplifiers generate significant heat loss, which burdens the cooling system and shortens the equipment's runtime.

[0003] Traditional power amplifier solutions have limitations in addressing such demands. For example, while Class A or Class AB linear power amplifiers can guarantee high signal linearity, their conduction losses are high, especially under light or half-load conditions, with efficiency typically below 60%, posing significant thermal management challenges in high-power applications. Class D power amplifiers, as a switching-mode amplifier, theoretically have near-100% efficiency, but traditional Class D power amplifiers based on silicon-based field-effect transistors suffer from significant switching losses under high-frequency, high-voltage operating conditions due to slow switching speeds and large parasitic parameters, making it difficult to achieve actual efficiencies exceeding 90% and resulting in low power density. Furthermore, the highly capacitive load characteristics of the transducer itself further exacerbate voltage stress and switching losses on the power devices, reducing system output stability and efficiency.

[0004] To overcome the limitations of traditional solutions, wide-bandgap semiconductor devices, such as gallium nitride (GaN), have been introduced into power stage designs. GaN devices feature high breakdown voltage, low on-resistance, and ultra-high switching speeds in the nanosecond range. When combined with Class D power amplifier topologies, they can improve efficiency, power density, and operating frequency, matching the driving requirements of sonar transducers.

[0005] However, current Class D power amplifier designs based on gallium nitride (GaN) devices still have shortcomings. On the one hand, some designs fail to fully utilize the high-frequency, high-voltage characteristics of GaN devices, resulting in insufficient optimization of switching losses and unrealized efficiency potential. On the other hand, the inherent nonlinearity and hysteresis effects of the transducer, as well as its electromechanical coupling characteristics, are sensitive to environmental factors such as temperature and pressure, leading to severe harmonic distortion in the drive signal. Especially in environments with fluctuating temperatures and pressures, such as the ocean, transducer characteristic drift exacerbates signal distortion, affecting the fidelity and reliability of the detection system. Existing power amplifier control methods mostly employ linear feedback or static compensation, which are insufficient to effectively suppress this complex, time-varying nonlinear distortion. Summary of the Invention

[0006] This invention provides a transducer driving device and method with neural network predistortion, which solves the problems of severe nonlinear distortion, low efficiency and poor environmental adaptability in the prior art when driving transducers.

[0007] To achieve the above objectives, the present invention provides a transducer driving device with neural network predistortion, comprising: a main control unit, a full-bridge power stage, a demodulator, and an isolation sampling circuit; The main control unit is characterized in that it includes a processing system and a programmable logic unit, and the main control unit is coupled with a temperature sensing module; The processing system runs a neural network predistortion algorithm, which is configured to receive a reference input signal, an actual output voltage signal fed back by the isolation sampling circuit, and real-time temperature data collected by the temperature sensing module to generate a predistortion compensation signal. The programmable logic unit is configured to generate a drive signal based on the predistortion compensation signal; The full-bridge power stage is configured to receive the drive signal, amplify the power, and then drive the transducer load through the demodulator. The isolation sampling circuit is configured to acquire the voltage signal at the output of the demodulator and feed it back to the processing system.

[0008] In a preferred embodiment, the neural network predistortion algorithm employs a gated recurrent unit architecture with residual connections; the neural network predistortion algorithm is further configured to receive pressure data and real-time impedance data, and use the reference input signal, the actual output voltage signal, the real-time temperature data, the pressure data, and the real-time impedance data as multi-dimensional feature inputs to generate the predistortion compensation signal.

[0009] In a preferred embodiment, the isolation sampling circuit includes a high-frequency current transformer for synchronously acquiring the input current signal of the transducer load; the processing system is configured to calculate the real-time impedance data based on the synchronously acquired actual output voltage signal and the input current signal.

[0010] In a preferred embodiment, the programmable logic unit is provided with a dynamic dead-time compensation engine; the neural network predistortion algorithm is further configured to generate a load phase prediction signal and output it to the dynamic dead-time compensation engine; the dynamic dead-time compensation engine is configured to adjust the dead time of the drive signal applied to the full-bridge power stage in real time according to the load phase prediction signal.

[0011] In a preferred embodiment, the full-bridge power stage is composed of gallium nitride high electron mobility transistors.

[0012] The present invention also provides a transducer driving method, characterized by comprising the following steps: The processing system runs a neural network predistortion algorithm, receiving a reference input signal, the actual output voltage signal fed back by the isolation sampling circuit, and real-time temperature data collected by the temperature sensing module; The neural network predistortion algorithm generates a predistortion compensation signal based on the received signal and data; The programmable logic unit generates a drive signal based on the pre-distortion compensation signal; The drive signal is received by the full-bridge power stage, amplified, and then driven by the demodulator to drive the transducer load. The voltage signal at the output of the demodulator is acquired by the isolation sampling circuit and fed back to the processing system.

[0013] In a preferred embodiment, the step of running the neural network predistortion algorithm specifically includes: performing computation using a gated recurrent unit architecture with residual connections; and also receiving pressure data and real-time impedance data, and using the reference input signal, the actual output voltage signal, the real-time temperature data, the pressure data, and the real-time impedance data as multi-dimensional feature inputs to generate the predistortion compensation signal.

[0014] In a preferred embodiment, before receiving the real-time impedance data, the method further includes the steps of: synchronously acquiring the input current signal of the transducer load through a high-frequency current transformer in the isolation sampling circuit; and calculating the real-time impedance data based on the synchronously acquired actual output voltage signal and the input current signal.

[0015] In a preferred embodiment, the step of generating a drive signal by the programmable logic unit based on the predistortion compensation signal further includes: generating a load phase prediction signal by the neural network predistortion algorithm; and adjusting the dead time of the drive signal in real time by a dynamic dead time compensation engine based on the load phase prediction signal.

[0016] In a preferred embodiment, the driving signal is a sinusoidal pulse width modulation signal.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. Reduce distortion and improve signal fidelity: By introducing a neural network predistortion algorithm, especially a gated cyclic unit architecture that can handle timing memory effects, and integrating multi-dimensional information such as temperature, pressure, and real-time impedance for adaptive compensation, it can model and counteract the inherent nonlinearity and hysteresis effects of the transducer and the characteristic drift caused by environmental factors in real time, thereby reducing the total harmonic distortion of the output signal.

[0018] 2. Improved efficiency and reliability: A full-bridge power stage is constructed using gallium nitride high electron mobility transistors, leveraging their high-frequency and high-efficiency characteristics. Combined with a dynamic dead-time compensation engine, the dead time of the power transistors is finely adjusted in real time using load phase information predicted by a neural network. This allows the power transistors to operate near the zero-voltage switching boundary, fundamentally suppressing switching losses and crossover distortion. It also avoids hard switching or reverse current caused by improper dead-time settings, thereby improving overall system efficiency and enhancing system reliability.

[0019] 3. Enhanced Environmental Adaptability: By using key environmental parameters such as temperature and pressure as inputs to the neural network model, the system achieves the perception and dynamic compensation of the transducer's nonlinear behavior under complex operating conditions. This enables the drive system to maintain stable and high-performance output even in harsh environments with drastic temperature and pressure variations, such as the deep sea and polar regions. It is particularly suitable for autonomous underwater platforms with extremely high environmental adaptability requirements, such as unmanned underwater vehicles and seabed observation stations. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0021] Figure 1 This is a system overall structure block diagram provided in the embodiments of the present invention.

[0022] Figure 2 This is a topology diagram of the subsequent circuit provided in an embodiment of the present invention.

[0023] Figure 3 This is a functional structure diagram of the main control unit provided in an embodiment of the present invention.

[0024] Figure 4 This is a schematic flowchart of the transducer driving method provided in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] It should be noted that in the description of this invention, the terms "coupling" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0027] Example 1 This embodiment provides a transducer driving device with neural network predistortion. This device can sense and dynamically compensate for the nonlinearity, hysteresis and temperature drift effects of the transducer in real time, so as to achieve high power and low distortion driving performance.

[0028] Please see Figure 1 , Figure 2 and Figure 3 , Figure 1 This is a system overall structure block diagram according to an embodiment of the present invention. Figure 2 This is a detailed topology diagram of the subsequent circuit. Figure 3 This is a functional block diagram of the main control unit. The transducer drive device mainly includes the main control unit, a full-bridge power stage, a demodulator, and an isolation sampling circuit. The system also includes a power supply module for providing a stable DC power supply to the full-bridge power stage, which can provide, for example, an adjustable DC bus voltage within ±350V.

[0029] The main control unit is the core of the entire system. In this embodiment, the main control unit adopts a heterogeneous computing architecture system-on-a-chip. For example... Figure 3 As shown, the system-on-a-chip integrates a processing system and a programmable logic unit.

[0030] The processing system integrates a dual-core processor and runs a lightweight operating system, responsible for executing complex control algorithms, system management, and communication with external host computers. The core of this invention, the neural network predistortion algorithm, runs on the processing system. The processing system receives feedback signals from the isolated sampling circuit and temperature data from the temperature sensing module. After noise filtering and data preprocessing, the data is fed into the neural network model for online inference to generate a predistortion compensation signal.

[0031] The programmable logic unit (PLU) utilizes its programmable logic resources to implement high-speed, real-time hardware logic. The PLU is responsible for generating the final drive signal based on the pre-distortion compensation signal generated by the processing system. Specifically, it internally contains a direct digital frequency synthesis structure, consisting of a phase accumulator and a sine lookup table, used to generate a high-precision sine wave reference. This reference signal is compared with a triangular wave carrier to generate the original sinusoidal pulse width modulation signal. Subsequently, this original signal is processed by a dead-time insertion module to prevent shoot-through between the upper and lower arms of the full-bridge power stage, and finally, a differential drive signal is output to the full-bridge power stage through the output register. The hardware logic response time of the PLU can reach the microsecond level, ensuring real-time control.

[0032] The main control unit is also coupled with a temperature sensing module. This module may include one or more high-precision digital temperature sensors, arranged in key locations such as the transducer housing and the full-bridge power stage heatsink, for real-time monitoring of the ambient or device operating temperature. The acquired real-time temperature data is synchronously input into a neural network predistortion algorithm running on the processing system, serving as an important input dimension of the model, enabling compensation to adapt to temperature changes.

[0033] The full-bridge power stage is the power amplification actuator. It receives drive signals generated from the programmable logic unit of the main control unit and converts them into high-frequency, high-voltage pulses sufficient to drive the transducers. For example... Figure 2 As shown, this full-bridge power stage consists of a full-bridge topology composed of four enhancement-mode gallium nitride (GaN) high-electron-mobility transistors (HETTs). Specifically, two intelligent power half-bridge chips can be used, each chip internally encapsulating a driver and a pair of GaN HETT half-bridges. Differential sinusoidal pulse-width modulation (PWM) signals from the programmable logic unit drive the two half-bridge chips respectively, controlling the high-speed switching of the four GaN HETTs, thereby generating a high-voltage pulse sequence with an amplitude equal to the DC bus voltage between the midpoints of the bridge arms. Thanks to the nanosecond-level switching speed and low on-resistance of GaN devices, this power stage can achieve an overall efficiency of over 90% with a ±350V supply and supports excitation frequencies up to 200kHz.

[0034] The demodulator's function is to filter out the high-frequency switching carrier components of the full-bridge power stage output, reconstructing a smooth analog drive voltage. In this embodiment, the demodulator uses a π-type LC low-pass filter. Figure 2 As shown, it consists of two series inductors L1 and L2 and two parallel capacitors C5 and C6. In a specific implementation, the inductance of inductors L1 and L2 can be 105 nH, and the capacitance of capacitors C5 and C6 can be 0.68 μF. The cutoff frequency of the filter is approximately 377 kHz, which can effectively filter out sinusoidal pulse width modulation carrier waves and their higher harmonics, while ensuring the integrity of the signal within the transducer's operating frequency band.

[0035] The isolation sampling circuit is responsible for safely and accurately feeding back the high-voltage analog signal applied to the transducer load at the demodulator output to the low-voltage digital domain of the main control unit. For example... Figure 2 As shown, the circuit first uses a voltage divider network composed of high-precision resistors R13, R14, R15, and R16 to proportionally attenuate the output voltage signal, which is as high as ±350V, to a lower range, such as within ±250 mV. The attenuated signal is then fed into a high-precision isolation amplifier. This amplifier utilizes capacitive coupling isolation technology to provide high common-mode transient immunity and high isolation withstand voltage, effectively blocking the interference of high voltage and noise from the power stage on the control circuit. After isolation amplification and conditioning, the output differential signal is scaled to a range acceptable to the analog-to-digital converter at the main control unit processing system, such as ±3.3V, serving as the actual output voltage signal for closed-loop feedback.

[0036] In a more preferred embodiment, to achieve deeper modeling and compensation of the nonlinear characteristics of the transducer under different operating conditions, this device has undergone functional enhancements based on the above scheme: The neural network algorithm has been upgraded, and the neural network predistortion algorithm running on the system side adopts a gated recurrent unit architecture with residual connections. Compared to standard gated recurrent units, residual connections allow certain layers of the model to directly learn identity mappings, which helps to solve the gradient vanishing problem during deep network training and can better capture linear components in the signal, making it suitable for handling the complex nonlinear-thermal coupled dynamic behavior of transducers. The state transition logic of its core computational unit can be described by the following formula: Reset door calculation:

[0037] Update gate calculation:

[0038] Candidate hidden state calculation:

[0039] Final hidden state calculation (including residual terms):

[0040] in, h t-1 This indicates the hidden state at the previous moment. x t This represents the input feature vector at the current time. r t and z t These represent the outputs of the reset gate and the update gate, respectively. ˜h t It is a candidate hidden state. h t It is the final hidden state at the current moment. σ tanh and tanh are activation functions. W and b These are the model's weights and bias parameters. The key lies in the residual term. F ( x t It directly passes a portion of the information from the original input to the next layer, preserving key temporal features.

[0041] In addition to the reference signal, feedback voltage, and temperature data, the algorithm is configured to receive additional pressure data and real-time impedance data as multi-dimensional feature inputs. These five types of data together constitute a multi-dimensional feature vector, which serves as the input to a gated cyclic unit model with residual connections. This enables the model to more comprehensively model the transducer nonlinearity under the coupling effects of multiple physical fields such as pressure, temperature, and impedance.

[0042] To obtain real-time impedance data, a high-frequency feedthrough current transformer is added to the isolation sampling circuit. This transformer is used to synchronously acquire the input current signal flowing through the transducer load, in conjunction with the voltage sampling. Based on the amplitude and phase information of the synchronously acquired voltage and current signals, the processing system can calculate the complex impedance of the transducer at the current operating frequency in real time.

[0043] Dynamic dead-time compensation replaces the traditional fixed dead-time insertion module at the programmable logic unit (PLU) level with a dynamic dead-time compensation engine. The gated loop unit model with residual connections at the processing system level generates a pre-distortion compensation signal and, leveraging its understanding of system dynamics, additionally generates a load phase prediction signal (variable). θ This signal is transmitted at high speed via the bus to the dynamic dead-time compensation engine at the programmable logic unit. Based on this phase prediction signal, the dynamic dead-time compensation engine adjusts the dead time of the drive signal in real time and with fine precision. t deadFor example, when the current phase is detected to lead the voltage phase, the dynamic dead-time compensation engine automatically reduces the default dead time (e.g., 40 ns) to a better value (e.g., 22 ns) in 2 ns increments to compensate for the slow change in switching node voltage caused by the transducer back electromotive force, thereby ensuring that gallium nitride high electron mobility transistors switch at the physical layer as close to zero voltage conditions as possible.

[0044] Through the aforementioned enhancement scheme, the intelligent prediction capability of the neural network extends from pure digital signal domain compensation to real-time control of the underlying physical switching characteristics of the power stage. This deeply coupled hardware and software collaborative mechanism can solve the problems of switching losses and harmonic distortion caused by drastic changes in dynamic impedance during frequency sweeping of piezoelectric loads. Experiments show that this scheme can reduce the total harmonic distortion under wideband frequency sweeping conditions from -35dB to below -55dB, while reducing the switching losses of the full-bridge power stage by approximately 15%.

[0045] Example 2 This embodiment provides a transducer driving method, which is implemented based on the driving device described in Embodiment 1. The specific execution flow is as follows: Figure 4 As shown.

[0046] Please see Figure 4 The transducer driving method S400 includes the following steps: Step S401: Multidimensional information collection and input.

[0047] After the system starts up, the processing system in the main control unit continuously receives multiple signals and data. These include: reference input signals provided by an external host computer or internal signal generator; actual output voltage signals representing the actual driving status of the transducers, fed back by the isolation sampling circuit; and real-time temperature data collected by the temperature sensing module, reflecting the system's operating environment and the status of key components.

[0048] Step S402: Run the neural network predistortion algorithm to generate a predistortion compensation signal.

[0049] The processing system runs a pre-trained neural network predistortion algorithm. This algorithm receives the signals and data acquired in step S401 as input. The neural network model models the dynamic behavior of the entire power amplification link and the transducer load. Based on the current input, the model infers and predicts the distortion that the system will produce online, and generates a forward-looking predistortion compensation signal accordingly. This signal is used to correct the reference input signal to compensate for the distortion that will occur in subsequent links.

[0050] Step S403: Generate drive signal.

[0051] The processing system transmits the generated predistortion compensation signal to the programmable logic unit (PLU). Based on this compensation signal, the PLU adjusts the parameters of its internal direct digital frequency synthesizer and pulse width modulation generator to generate an optimized drive signal. This drive signal is typically a sinusoidal pulse width modulation signal. During the generation process, a necessary dead time is also inserted to protect the full-bridge power stage.

[0052] Step S404: Power amplification and driving.

[0053] The drive signal generated by the programmable logic unit is sent to the full-bridge power stage. The full-bridge power stage, composed of gallium nitride high electron mobility transistors, receives the drive signal, amplifies it, and outputs a high-frequency, high-voltage pulse current.

[0054] Step S405: Demodulation and load driving.

[0055] The amplified signal is filtered by a demodulator to remove high-frequency carrier components, reconstructing a smooth analog drive voltage, which is then applied to the transducer load to drive it to work.

[0056] Step S406: Closed-loop feedback sampling.

[0057] The isolation sampling circuit located at the output of the demodulator collects the actual voltage signal applied to the transducer in real time. After isolation, attenuation and conditioning, it is fed back to the processing system as the actual output voltage signal for step S401, thus forming a complete closed-loop control system.

[0058] In a preferred embodiment of the method, to further improve driving performance and environmental adaptability, the above steps can be refined and enhanced as follows: In step S401, in addition to receiving voltage, temperature, and reference signals, pressure data provided by a pressure sensor is also received. Simultaneously, the input current signal flowing through the transducer is synchronously acquired via a high-frequency current transformer in the isolation sampling circuit. The processing system calculates real-time impedance data based on the synchronously acquired voltage and current signals. These five data types (reference signal, feedback voltage, temperature, pressure, and impedance) together constitute a multi-dimensional feature input.

[0059] In step S402, the neural network predistortion algorithm is specifically operated using a gated recurrent unit architecture with residual connections. This architecture utilizes its temporal modeling capabilities to process the aforementioned multidimensional feature inputs, thereby generating a more accurate predistortion compensation signal.

[0060] In step S403, this step further includes a sub-step of dynamic dead-time compensation. Specifically, in step S402, the gated loop unit algorithm with residual connections generates a load phase prediction signal while generating the pre-distortion compensation signal. In step S403, this phase prediction signal is passed to the dynamic dead-time compensation engine within the programmable logic unit. The dynamic dead-time compensation engine adjusts the dead time inserted into the drive signal in real time and dynamically based on the prediction signal. This step tightly integrates the predictive capability of the algorithm with the physical control of the hardware, fundamentally optimizing the switching process and reducing losses and distortion.

[0061] In summary, this invention combines advanced neural network algorithms with high-performance gallium nitride power electronics technology and introduces a multi-physics sensing and hardware / software co-control mechanism to achieve a high-efficiency, high-fidelity, and highly environmentally adaptable transducer driving scheme.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A transducer driving device with neural network predistortion, comprising a main control unit, a full-bridge power stage, a demodulator, and an isolation sampling circuit, characterized in that: The main control unit includes a processing system and a programmable logic unit, and the main control unit is coupled with a temperature sensing module; The processing system runs a neural network predistortion algorithm, which is configured to receive a reference input signal, an actual output voltage signal fed back by the isolation sampling circuit, and real-time temperature data collected by the temperature sensing module to generate a predistortion compensation signal. The programmable logic unit is configured to generate a drive signal based on the predistortion compensation signal; The full-bridge power stage is configured to receive the drive signal, amplify the power, and then drive the transducer load through the demodulator. The isolation sampling circuit is configured to acquire the voltage signal at the output of the demodulator and feed it back to the processing system.

2. The transducer drive device according to claim 1, characterized in that, The neural network predistortion algorithm adopts a gated recurrent unit architecture with residual connections; the neural network predistortion algorithm is also configured to receive pressure data and real-time impedance data, and use the reference input signal, the actual output voltage signal, the real-time temperature data, the pressure data and the real-time impedance data as multi-dimensional feature inputs to generate the predistortion compensation signal.

3. The transducer drive device according to claim 2, characterized in that, The isolation sampling circuit includes a high-frequency current transformer, which is used to synchronously acquire the input current signal of the transducer load; the processing system is configured to calculate the real-time impedance data based on the synchronously acquired actual output voltage signal and the input current signal.

4. The transducer drive device according to claim 1, characterized in that, The programmable logic unit is equipped with a dynamic dead-time compensation engine; the neural network predistortion algorithm is further configured to generate a load phase prediction signal and output it to the dynamic dead-time compensation engine; the dynamic dead-time compensation engine is configured to adjust the dead time of the drive signal applied to the full-bridge power stage in real time according to the load phase prediction signal.

5. The transducer drive device according to claim 1, characterized in that, The full-bridge power stage is composed of gallium nitride high electron mobility transistors.

6. A transducer driving method, characterized in that, Includes the following steps: The neural network predistortion algorithm is run to receive the reference input signal, the actual output voltage signal fed back by the isolation sampling circuit, and the real-time temperature data collected by the temperature sensing module. Based on the received signals and data, a predistortion compensation signal is generated; A driving signal is generated based on the pre-distortion compensation signal; The drive signal is received, amplified, and then used to drive the transducer load via a demodulator. The voltage signal at the output of the demodulator is acquired and fed back to the processing system.

7. The transducer driving method according to claim 6, characterized in that, The steps of running the neural network predistortion algorithm specifically include: performing computation using a gated recurrent unit architecture with residual connections; and receiving pressure data and real-time impedance data, and using the reference input signal, the actual output voltage signal, the real-time temperature data, the pressure data, and the real-time impedance data as multi-dimensional feature inputs to generate the predistortion compensation signal.

8. The transducer driving method according to claim 7, characterized in that, Before receiving the real-time impedance data, the method further includes the step of: synchronously acquiring the input current signal of the transducer load through the high-frequency current transformer in the isolation sampling circuit. The real-time impedance data is calculated based on the synchronously acquired actual output voltage signal and input current signal.

9. The transducer driving method according to claim 6, characterized in that, The step of generating the drive signal further includes: generating an additional load phase prediction signal; and adjusting the dead time of the drive signal in real time using a dynamic dead time compensation engine based on the load phase prediction signal.

10. The transducer driving method according to claim 6, characterized in that, The driving signal is a sinusoidal pulse width modulation signal.

Images