A noise reduction system and method based on a sequence pipeline transposed RDFxLMS

Abstract

The application relates to a noise reduction system and method based on a sequence pipeline transposed RDFxLMS, wherein a FIR filter and a secondary path adopt a transposed form, the key path in the algorithm structure is short, and the clock speed of the system can be greatly improved. In the algorithm, an input signal is filtered through a FIR filter module to obtain a filter output. An error calculation module mainly comprises a transposed FIR filter and a subtractor, and is used for calculating an error signal between the filter output and an expected output. A weight value updating module calculates new weight values according to the error signal and the filter input, and stores the new weight values in a weight value memory. A secondary path module is used for correcting an error gradient estimation value of the LMS algorithm, so that the convergence performance of the filter is improved. A secondary path output is calculated according to the new weight values and the filter input, and the secondary path output is added to a main path output to obtain a final output signal. The secondary path output signal is used for correcting the weight values of the main path, so that the noise reduction effect is improved.

Description

Technical Field

[0001] This invention belongs to the field of digital signal processing, specifically relating to a design method for a sequence pipelined transpose RDFxLMS algorithm. Background Technology

[0002] With rapid economic development and urbanization, noise pollution seriously threatens people's physical and mental health, making noise control an urgent problem to solve. Noise control methods can be divided into passive noise control and active noise control. Active noise cancellation (ANC) is widely used in active noise-canceling headphones because it effectively reduces low-frequency noise. This invention focuses on the research and structural improvement of active noise control algorithms, while also considering the analysis of adaptive filter structures that consider power consumption, algorithm convergence, and throughput. In practical applications, ensuring the convergence of the adaptive filter while reducing power consumption and increasing throughput are crucial factors for hardware implementation. Therefore, this invention investigates the feasibility of implementing the FxLMS algorithm in active noise-canceling headphones and proposes a sequence pipelined transpose RDFxLMS algorithm system and method.

[0003] The current challenge in improving the performance of active noise control systems lies in ensuring the convergence of adaptive filters while reducing power consumption and increasing throughput. The traditional FxLMS algorithm uses an FIR filter to filter input noise and employs the LMS algorithm to update the weights of the control filter. However, the dispersion of eigenvalues ​​in the autocorrelation matrix of the input signal leads to a slow convergence mode in the LMS algorithm, making it impossible to achieve compatibility between convergence speed, power consumption, and throughput. To address this, we propose a sequence pipelined transposed RDFxLMS algorithm system. In this system, the FIR filter and secondary path are transposed, resulting in a shorter critical path. This significantly improves the system's clock speed, leading to faster convergence and lower power consumption. Simultaneously, the transposed path also increases data throughput, effectively improving the performance of the active noise control system. Summary of the Invention

[0004] This invention aims to solve the problems of the prior art mentioned above. It proposes a noise reduction system and method based on sequence pipelined transposed RDFxLMS. The technical solution of this invention is as follows:

[0005] A noise reduction system based on sequence pipelined transposed RDFxLMS, ​​comprising:

[0006] The system comprises an FIR filter module, an error calculation module, a weight update module, and a secondary path module. The FIR filter module filters out noise signals outside the signal bandwidth from the input signal to obtain the filter output signal. The FIR filter is a fully pipelined structure based on sequence delay LMS. By retiming the inserted delay to compensate for the delay of the MAC and upsampler, a sequence retiming delay LMS adaptive FIR filter is obtained. One delay in the original signal is equivalent to N delays in the upsampled signal.

[0007] The error calculation module mainly consists of a transposed FIR filter and a subtractor. The transposed FIR filter is used to filter the desired output signal, and the subtractor is used to calculate the error signal between the filter output and the desired output.

[0008] The weight update module is used to calculate new weights based on the error signal and the filter output signal, and store the new weights in the weight memory; the convergence factor in the weight update module is a negative integer power of 2, which is achieved through corresponding shift operations;

[0009] The secondary path module is used to correct the error gradient estimate of the LMS algorithm. It calculates the secondary path output based on the new weights and filter input, and adds it to the main path output to obtain the final output signal. The secondary path output signal is used to correct the weight update of the main path, so that the error gradient of the main path can be estimated more accurately and the noise reduction effect can be improved.

[0010] Furthermore, the error calculation module is used to calculate the error signal between the filter output and the desired output, specifically including:

[0011] The filter output and the desired output are processed by the error signal module. The filter output is then passed through a transposed FIR filter and subtracted from the desired output to obtain the error signal as shown in formula (1):

[0012] (1)

[0013] In the formula, The main noise signal after adding adaptive delay; It is a secondary path estimation signal; It is the filtered output signal; This refers to the input noise signal; This represents the transfer function of the transposed FIR filter; This refers to the input noise signal after passing through the secondary path.

[0014] Furthermore, the weight update module is used to calculate new weights based on the error signal and the filter output signal, specifically including:

[0015] By comparing the error signal with the input filter signal, the weight coefficients of the transverse filter and the reference input signal at time n are calculated, as shown in formulas (2) and (3) respectively:

[0016] (2)

[0017] (3)

[0018] The weight coefficient obtained in the nth iteration. The secondary path module, representing the reference input signal at time n-L+2, is used to correct the error gradient estimate of the LMS algorithm. It calculates the secondary path output based on the new weights and filter input, and adds it to the main path output to obtain the final output signal. Specifically, it includes:

[0019] The obtained weight coefficients are imported into the secondary path to correct the error gradient estimate of the LMS algorithm and remove the error. The secondary path rewrites formula (1) according to the weight update, as shown in formula (4):

[0020] (4)

[0021] i represents the data address, N represents the delay number, and the filter coefficients are recursively derived according to the steepest descent method principle. The instantaneous mean square error is used to replace the mean square error calculation, as shown in formula (5):

[0022] (5)

[0023] The gradient is represented as shown in equation (6):

[0024] (6)

[0025] Substituting equation (6) into equation (5), we get equation (7):

[0026] (7)

[0027] When the tap length of the adaptive filter is long enough, the step size limit of the transpose RDFxLMS algorithm is as shown in formula (8):

[0028] (8)

[0029] In the formula, The step size factor must satisfy the condition of the largest eigenvalue; This represents the maximum eigenvalue of the transposed RDFxLMS algorithm.

[0030] A noise reduction method based on any one of the systems described above, comprising the following steps:

[0031] Step 1: Reset the entire circuit to clear the stored values ​​in the weight update module;

[0032] Step 2: After the algorithm starts running, the input noise signal is initially filtered by the FIR filter module to obtain the filter output;

[0033] Step 3: The filter output and the desired output are processed by the error signal module. The filter output is passed through a transposed FIR filter and then subtracted from the desired output to obtain the error signal.

[0034] Step four: Compare the acquired error signal with the signal of the input filter, calculate the new weights, store them in the weight memory, and calculate the new reference input signal;

[0035] Step 5: Import the obtained weights into the secondary path to correct the error gradient estimate of the LMS algorithm and remove the error.

[0036] Step 6: Calculate the secondary path output based on the new weights and filter input, and add it to the primary path output to obtain the final output signal.

[0037] Furthermore, in step three, the error signal obtained through the FIR filter and subtractor of the error processing module is shown in formula (1):

[0038] (1)

[0039] In the formula, The main noise signal after adding adaptive delay; It is a secondary path estimation signal; It is the filtered output signal; This refers to the input noise signal; This represents the transfer function of the transposed FIR filter; This refers to the input noise signal after passing through the secondary path.

[0040] Furthermore, by comparing the error signal with the input filter signal, the weight coefficients of the transverse filter at time n and the reference input signal are calculated, as shown in formulas (2) and (3) respectively:

[0041] (2)

[0042] (3)

[0043] The weight coefficient obtained in the nth iteration. This represents the reference input signal at time n-L+2. Further, step five specifically includes: the secondary path rewrites formula (1) based on the weight update, as shown in formula (4):

[0044] (4)

[0045] i represents the data address, N represents the delay number, and the filter coefficients are recursively derived according to the steepest descent method principle. The instantaneous mean square error is used to replace the mean square error calculation, as shown in formula (5):

[0046] (5)

[0047] The gradient is represented as shown in equation (6):

[0048] (6)

[0049] Substituting equation (6) into equation (5), we get equation (7):

[0050] (7)

[0051] When the tap length of the adaptive filter is long enough, the step size limit of the transpose RDFxLMS algorithm is as shown in formula (8):

[0052] (8)

[0053] In the formula, The step size factor must satisfy the condition of the largest eigenvalue; This represents the maximum eigenvalue of the transposed RDFxLMS algorithm.

[0054] Furthermore, the adaptive FIR filter is a fully pipelined structure based on sequence delay LMS. By retiming the inserted delay to compensate for the delay of the MAC and the upsampler, a sequence retiming delay LMS adaptive FIR filter is obtained. One delay in the original signal is equivalent to N delays in the upsampled signal.

[0055] Furthermore, the multiply-accumulate-add unit is a simple architecture that uses a multiplier and an accumulator sequentially to implement the FIR filter; for the input data Upsampling is performed to improve the filter output obtained from downsampling. Sequential calculations are performed; input data and filter coefficients are stored using a dual-port block RAM, which will be used for mixed-mode configuration. Data is written to and read from port A (RAM mode), and coefficients are read from port B (ROM mode). For a given data address i, a coefficient address N + i is taken, and the MAC is multiplied by the corresponding delayed data sample and accumulated to obtain the result. The filter output is available after N address clock cycles.

[0056] The advantages and beneficial effects of this invention are as follows:

[0057] This invention relates to a design method for a sequence pipelined transposed RDFxLMS algorithm. In this algorithm design method, the FIR filter is transposed, which shortens the critical path in the algorithm structure and greatly improves the clock speed of the system. In addition, the algorithm optimizes the delay aspect. One delay (Z−1) in the original signal is equivalent to N delays (Z−N) in the upsampled signal. These advantages effectively improve the noise reduction performance in the active noise control process.

[0058] A block diagram of the transposed RDFxLMS algorithm of the present invention is shown below. Figure 1 As shown in the diagram, the algorithm's structure includes an FIR filter module, an error calculation module, a weight update module, and a secondary path module. The adaptive delay is 2, which ensures that the algorithm can effectively reduce noise.

[0059] The algorithm first inputs the input signal into an FIR filter, then obtains the error signal through an error calculation module. The weight update module updates the weights based on the error signal and a reference signal, providing new weights for the error gradient estimate of the secondary path correction LMS algorithm, thereby improving the filter's convergence performance. Then, the secondary path output is calculated based on the new weights and the filter input, and added to the primary path output to obtain the final output signal.

[0060] The main innovation of this invention lies in the combination of steps three to five, and the addition of a sequence retiming delay LMS adaptive FIR filter (SF-TF-RDFxLMS). Steps three to five, by adding an error calculation module, change the original method of providing output based on the acquired signal. By obtaining the difference between the real-time output and the ideal output through the error calculation module, the calculation weights of the output signal are adjusted, realizing adjustment based on the real-time output signal, which can effectively improve the noise reduction performance of the system. By adding SF-TF-RDFxLMS, ​​a sequence pipelined transpose RDFxLMS algorithm is implemented. This architecture can improve the operating frequency by minimizing the critical path, and the architecture is optimized in terms of delay. One delay (Z−1) in the original signal is equivalent to N delays (Z−N) in the upsampled signal. Attached Figure Description

[0061] Figure 1 This is a block diagram of the transposed RDFxLMS algorithm structure of the present invention, provided in a preferred embodiment.

[0062] Figure 2 This is a block diagram of the sequence pipelined transpose RDFxLMS algorithm structure of the present invention;

[0063] Figure 3 This section compares the convergence of the sequence pipeline transpose RDFxLMS algorithm. Detailed Implementation

[0064] The technical solutions of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0065] The technical solution of the present invention to solve the above-mentioned technical problems is:

[0066] Figure 1 The diagram shows the block structure of the transposed RDFxLMS algorithm. A noise reduction system based on sequence pipelined transposed RDFxLMS includes:

[0067] The system comprises an FIR filter module, an error calculation module, a weight update module, and a secondary path module. The FIR filter module filters out noise signals outside the signal bandwidth from the input signal to obtain the filter output signal. The FIR filter is a fully pipelined structure based on sequence delay LMS. By retiming the inserted delay to compensate for the delay of the MAC and upsampler, a sequence retiming delay LMS adaptive FIR filter is obtained. One delay in the original signal is equivalent to N delays in the upsampled signal.

[0068] The error calculation module mainly consists of a transposed FIR filter and a subtractor, and is used to calculate the error signal between the filter output and the desired output.

[0069] The weight update module is used to calculate new weights based on the error signal and the filter, and store the new weights in the weight memory. The convergence factor is a negative integer power of 2, which is achieved through corresponding shift operations.

[0070] The secondary path module is used to correct the error gradient estimate of the LMS algorithm. It calculates the secondary path output based on the new weights and filter input, and adds it to the main path output to obtain the final output signal. The secondary path output signal is used to correct the weight update of the main path, so that the error gradient of the main path can be estimated more accurately and the noise reduction effect can be improved.

[0071] Preferably, the error calculation module is used to calculate the error signal between the filter output and the desired output, specifically including:

[0072] The filter output and the desired output are processed by the error signal module. The filter output is then passed through a transposed FIR filter and subtracted from the desired output to obtain the error signal as shown in formula (1):

[0073] (1)

[0074] In the formula, The main noise signal after adding adaptive delay; It is a secondary path estimation signal; It is the filtered output signal; It is the filtered output signal; This refers to the input noise signal; This represents the transfer function of the transposed FIR filter; This refers to the input noise signal after passing through the secondary path.

[0075] Preferably, the weight update module is used to calculate new weights based on the error signal and the filter, specifically including:

[0076] By comparing the error signal with the input filter signal, the weight coefficients of the transverse filter and the reference input signal at time n are calculated, as shown in formulas (2) and (3) respectively:

[0077] (2)

[0078] (3)

[0079] The weight coefficient obtained in the nth iteration. This represents the reference input signal at time n-L+2.

[0080] Preferably, the secondary path module is used to correct the error gradient estimate of the LMS algorithm, calculate the secondary path output based on the new weights and filter input, and add it to the main path output to obtain the final output signal, specifically including:

[0081] The obtained weight coefficients are imported into the secondary path to correct the error gradient estimate of the LMS algorithm and remove the error. The secondary path rewrites formula (1) according to the weight update, as shown in formula (4):

[0082] (4)

[0083] i represents the data address, N represents the delay number, and the filter coefficients are recursively derived according to the steepest descent method principle. The instantaneous mean square error is used to replace the mean square error calculation, as shown in formula (5):

[0084] (5)

[0085] The gradient is represented as shown in equation (6):

[0086] (6)

[0087] Substituting equation (6) into equation (5), we get equation (7):

[0088] (7)

[0089] When the tap length of the adaptive filter is long enough, the step size limit of the transpose RDFxLMS algorithm is as shown in formula (8):

[0090] (8)

[0091] In the formula, The step size factor must satisfy the condition of the largest eigenvalue; This represents the maximum eigenvalue of the transposed RDFxLMS algorithm.

[0092] This explanation uses the processing of a noise signal as an example. The main steps are as follows:

[0093] Step 1: Reset the entire circuit to clear the stored weight values ​​in the weight update module to zero;

[0094] Step 2: After the algorithm starts running, the input noise signal is initially filtered by the FIR filter module to obtain the filter output;

[0095] Step 3: Process the filter output and the desired output through the error signal module. The filter output is subtracted from the desired output after passing through the transposed FIR filter to obtain the error signal as shown in formula (1):

[0096] (1)

[0097] In the formula, The main noise signal after adding adaptive delay; It is a secondary path estimation signal; It is the output signal after filtering.

[0098] Step four: Compare the acquired error signal with the signal of the input filter, calculate the new weight coefficients, store them in the weight memory, and calculate the new reference input signal. The weight coefficients of the transverse filter and the reference input signal at time n are shown in formula (2) and formula (3) respectively:

[0099] (2)

[0100] (3)

[0101] Step 5: Import the obtained weight coefficients into the secondary path to correct the error gradient estimate of the LMS algorithm and remove the error. The secondary path rewrites formula (1) according to the weight update, as shown in formula (4):

[0102] (4)

[0103] The filter coefficients are recursively derived based on the steepest descent method principle, and the instantaneous mean square error is used instead of the mean square error, as shown in formula (5):

[0104] (5)

[0105] The gradient is represented as shown in equation (6):

[0106] (6)

[0107] Substituting equation (6) into equation (5), we get equation (7):

[0108] (7)

[0109] When the tap length of the adaptive filter is long enough, the step size limit of the transpose RDFxLMS algorithm is as shown in formula (8):

[0110] (8)

[0111] In the formula, is the step size factor that must satisfy the maximum eigenvalue; is the maximum eigenvalue of the transposed RDFxLMS algorithm.

[0112] Step 6: Calculate the secondary path output based on the new weights and filter input, and add it to the primary path output to obtain the final output signal.

[0113] Figure 3 This section compares the convergence of the sequence pipeline transpose RDFxLMS algorithm.

[0114] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions.

[0115] It should also be noted that 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 process, method, article, or apparatus. Without further limitation, 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.

[0116] The above embodiments should be understood as illustrative only and not as limiting the scope of protection of the present invention. After reading the description of the present invention, those skilled in the art can make various alterations or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A noise reduction system based on sequence pipelined transposed RDFxLMS, ​​characterized in that, include: The system comprises an FIR filter module, an error calculation module, a weight update module, and a secondary path module. The FIR filter module filters out noise signals outside the signal bandwidth from the input signal to obtain the filter output signal. The FIR filter is a fully pipelined structure based on sequence delay LMS. By retiming the inserted delay to compensate for the delay of the MAC and upsampler, a sequence retiming delay LMS adaptive FIR filter is obtained. One delay in the original signal is equivalent to N delays in the upsampled signal. The error calculation module mainly consists of a transposed FIR filter and a subtractor. The transposed FIR filter is used to filter the desired output signal, and the subtractor is used to calculate the error signal between the filter output and the desired output. The weight update module is used to calculate new weights based on the error signal and the filter output signal, and store the new weights in the weight memory; the convergence factor in the weight update module is a negative integer power of 2, which is achieved through corresponding shift operations; The secondary path module is used to correct the error gradient estimate of the LMS algorithm. It calculates the secondary path output based on the new weights and filter input, and adds it to the main path output to obtain the final output signal. The secondary path output signal is used to correct the weight update of the main path, so as to more accurately estimate the error gradient of the main path and improve the noise reduction effect. The FIR filter module includes a multiply-accumulate-address unit, which implements the FIR filter sequentially using a multiplier and an accumulator; for the input data... Upsampling is performed to improve the filter output obtained from downsampling. Sequential calculations are performed; input data and filter coefficients are stored using dual-port RAM, which will be used for mixed-mode configuration. Data is written to and read from port A, and coefficients are read from port B. For a given data address i, a coefficient address N + i is taken, and the MAC is multiplied by the corresponding delayed data sample and accumulated to obtain the result. The filter output is available after N address clock cycles.

2. The noise reduction system based on sequence pipelined transposed RDFxLMS according to claim 1, characterized in that, The error calculation module is used to calculate the error signal between the filter output and the desired output, specifically including: The filter output and the desired output are processed by the error signal module. The filter output is then passed through a transposed FIR filter and subtracted from the desired output to obtain the error signal as shown in formula (1): （1） In the formula, This is to incorporate the main noise signal after adding adaptive delay; It is a secondary path estimation signal; It is the filtered output signal; This refers to the input noise signal; This represents the transfer function of the transposed FIR filter; This refers to the input noise signal after passing through the secondary path.

3. The noise reduction system based on sequence pipelined transposed RDFxLMS according to claim 2, characterized in that, The weight update module is used to calculate new weights based on the error signal and the filter output signal, specifically including: By comparing the error signal with the input filter signal, the weight coefficients of the transverse filter and the reference input signal at time n are calculated, as shown in formulas (2) and (3) respectively: （2） （3） The weight coefficient obtained in the nth iteration. This represents the reference input signal at time n-L+2.

4. A noise reduction system based on sequence pipelined transposed RDFxLMS according to claim 2, characterized in that, The secondary path module is used to correct the error gradient estimate of the LMS algorithm. It calculates the secondary path output based on the new weights and filter input, and adds it to the main path output to obtain the final output signal. Specifically, it includes: The obtained weight coefficients are imported into the secondary path to correct the error gradient estimate of the LMS algorithm and remove the error. The secondary path rewrites formula (1) according to the weight update, as shown in formula (4): （4） i represents the data address, N represents the delay number, and the filter coefficients are recursively derived according to the steepest descent method principle. The instantaneous mean square error is used instead of the mean square error, as shown in formula (5): （5） The gradient is represented as shown in equation (6): （6） Substituting equation (6) into equation (5), we get equation (7): （7） When the tap length of the adaptive filter is long enough, the step size limit of the transpose RDFxLMS algorithm is as shown in formula (8): （8） In the formula, The step size factor must satisfy the condition of the largest eigenvalue; This represents the maximum eigenvalue of the transposed RDFxLMS algorithm.

5. A noise reduction method based on the system according to any one of claims 1-4, characterized in that, Includes the following steps: Step 1: Reset the entire circuit to clear the stored values ​​in the weight update module; Step 2: After the algorithm starts running, the input noise signal is initially filtered by the FIR filter module to obtain the filter output; Step 3: The filter output and the desired output are processed by the error signal module. The filter output is subtracted from the desired output after passing through the transposed FIR filter to obtain the error signal. Step four: Compare the acquired error signal with the signal of the input filter, calculate the new weights, store them in the weight memory, and calculate the new reference input signal; Step 5: Import the obtained weights into the secondary path to correct the error gradient estimate of the LMS algorithm and remove the error. Step 6: Calculate the secondary path output based on the new weights and filter input, and add it to the primary path output to obtain the final output signal.

6. The noise reduction method according to claim 5, characterized in that, In step three, the error signal obtained through the FIR filter and subtractor of the error processing module is shown in formula (1): （1） In the formula, This is to incorporate the main noise signal after adding adaptive delay; It is a secondary path estimation signal; It is the filtered output signal; This refers to the input noise signal; This represents the transfer function of the transposed FIR filter; This refers to the input noise signal after passing through the secondary path.

7. The noise reduction method according to claim 5, characterized in that, Step four specifically includes: calculating the weight coefficients of the transverse filter and the reference input signal at time n by comparing the error signal and the input filter signal, as shown in formulas (2) and (3) respectively: （2） （3） The weight coefficient obtained in the nth iteration. This represents the reference input signal at time n-L+2.

8. The noise reduction method according to claim 5, characterized in that, Step five specifically includes: rewriting formula (1) based on the weight update of the secondary path, as shown in formula (4): （4） i represents the data address, N represents the delay number, and the filter coefficients are recursively derived according to the steepest descent method principle. The instantaneous mean square error is used instead of the mean square error, as shown in formula (5): （5） The gradient is represented as shown in equation (6): （6） Substituting equation (6) into equation (5), we get equation (7): （7） When the tap length of the adaptive filter is long enough, the step size limit of the transpose RDFxLMS algorithm is as shown in formula (8): （8） In the formula, The step size factor must satisfy the condition of the largest eigenvalue; This represents the maximum eigenvalue of the transposed RDFxLMS algorithm.

9. The noise reduction method according to claim 5, characterized in that, The adaptive FIR filter is a fully pipelined structure based on sequence delay LMS. By retiming the inserted delay to compensate for the delay of the MAC and the upsampler, a sequence retiming delay LMS adaptive FIR filter is obtained. One delay in the original signal is equivalent to N delays in the upsampled signal.

Images