Digital super-resolution low-bit quantization method and system based on multi-offset observation fusion

Abstract

The invention discloses a digital super-resolution low-bit quantization method and system based on multi-offset observation fusion, and the method comprises the following steps: 1, carrying out the amplitude normalization processing of a to-be-quantized signal according to the amplitude range of the signal, the input dynamic range of a uniform quantizer, and a quantization step length LSB; 2, a centrosymmetric and equidistant jitter sequence is constructed based on the quantization step length LSB, and the value interval of the jitter sequence covers [-LSB / 2, LSB / 2]; 3, respectively superposing each jitter sample on the normalized signal to form a plurality of groups of disturbed signals, and uniformly quantizing each group of signals under the same quantization threshold structure; 4, averaging the obtained multiple groups of quantization results in a sample dimension, and constructing an output signal sequence with reduced equivalent quantization error variance; and step 5, carrying out inverse transformation processing on the obtained output signal to enable the output signal to return to an original amplitude space.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of optical communication, and relates to a low-bit precision non-uniform quantization method of signals in a multi-carrier system, in particular to a digital super-resolution low-bit quantization method and system based on multi-offset observation fusion. BACKGROUND

[0002] With the deep integration of high-speed optical communication and wireless communication, optoelectronic hybrid transmission systems are becoming an important direction for future information transmission. Data center interconnection, optical access network and short-distance high-speed interconnection scenarios have higher requirements for speed, bandwidth density and energy efficiency. Traditional electrical transmission is limited by bandwidth, power consumption and distance, while optical communication provides a new idea for building efficient and low-latency networks due to its ultra-large bandwidth, low loss and strong anti-interference capability. At the same time, the development of emerging technologies such as 6G, terahertz and visible light communication has also accelerated the evolution of optoelectronic hybrid systems. By realizing flexible signal processing in the electrical domain and high-speed transmission in the optical domain, optoelectronic hybrid systems have the advantages of both and have become a key technology for realizing high-bandwidth, low-latency and low-power interconnection.

[0003] The current mainstream optical communication adopts the intensity modulation direct detection (IMDD) architecture, which is simple in structure, low in power consumption and easy to integrate, and has become the main technical route for new-generation data center and computing power network high-speed optical interconnection. Under the promotion of digital signal processing, orthogonal frequency division multiplexing (OFDM) has become the core modulation method for improving the performance and intelligent level of IMDD systems, with high spectral efficiency, anti-multipath characteristics and adaptive configuration, combined with machine learning equalization and compensation technology. However, IMDD-OFDM systems still face two major challenges in high-speed real-time implementation: one is the severe distortion caused by the nonlinear response of optoelectronic devices and high peak-to-average ratio (PAPR); the other is the significant increase in power consumption and hardware complexity caused by high-precision ADC / DAC. Although high-precision ADC / DAC can effectively reduce quantization noise, improve error vector magnitude (EVM) and bit error rate (BER), its power consumption, chip area and complexity increase exponentially with resolution and sampling rate, which seriously restricts the energy efficiency and integration of the system. Therefore, the research on low-precision, high-energy-efficiency ADC / DAC schemes and how to balance performance and complexity has become a core problem in the design of IMDD-OFDM systems.

[0004] Low-bit quantization schemes are concerned for their low power consumption, simple structure and easy integration, but the introduced quantization noise and nonlinear distortion will significantly affect the signal quality. Therefore, domestic and foreign scholars have proposed various optimization strategies, mainly including digital resolution enhancement, noise shaping, statistically characteristic driven non-uniform quantization and optimal target oriented quantization design. Bi et al. and Lin et al. estimate the signal distribution based on non-parametric histogram to optimize the quantization level, which significantly improves the BER and sensitivity under 3-5 bit quantization, but the calculation accuracy of this scheme depends heavily on the improvement of calculation complexity and has low flexibility in implementation and deployment. Xu et al. directly estimate the optimal quantization threshold based on Gaussian mixture model (GMM) and expectation maximization (EM) algorithm, which improves the EVM by 2.7%-7.8% and the receiving sensitivity by about 1 dB under 3-5 bit quantization, but the random initialization of GMM initial parameters may affect the stability and optimality of the quantization level. Based on this, the present application proposes a digital super-resolution quantization method and system based on multi-offset observation fusion for high bandwidth and high real-time optical communication systems such as IMDD-OFDM. SUMMARY

[0005] In view of the key problems of the existing low-precision and high-fidelity ADC / DAC schemes and the balance between performance and complexity, the present application provides a digital super-resolution low-bit quantization method and system based on multi-offset observation fusion. The dithering signal is a small amplitude noise signal superimposed on the original signal, which can make the quantized signal and the quantization noise irrelevant, thereby reducing the distortion of the quantized signal. By establishing a mathematical model and analyzing the quantization noise and system performance, the present application can greatly improve the quantization quality of the signal under low-bit precision conditions by adding the generated specific dithering sequence to the OFDM signal to be quantized, and realize low-cost and high-performance DAC.

[0006] To achieve the above purpose, the present application adopts the following technical solutions: A digital super-resolution low-bit quantization method based on multi-offset observation fusion, the specific steps are as follows: Step 1: According to the amplitude range of the signal and the input dynamic range and quantization step LSB of uniform quantization, the amplitude of the signal to be quantized is normalized to realize the consistency of the signal and the scale of the quantizer; Step 2: According to the quantization step LSB, a specific dithering sequence with central symmetry and equal interval is constructed, covering the range [-LSB / 2, LSB / 2], which is used to introduce controllable disturbance in the subsequent parallel quantization processing; Step 3: After scrambling each dithering sequence, the same threshold structure is used for quantization to realize multiple independent quantization sampling; Step 4: average the quantization results of step 3 by sample dimension to obtain an output signal with reduced equivalent error variance; Step 5: perform inverse transform processing on the quantization results of step 4 to restore the signal to the original amplitude space.

[0007] Preferably, in step 1, according to the amplitude range of the time domain signal x(t) and the quantization step LSB of the uniform quantizer, amplitude normalization is performed on the signal to be quantized x(t) to obtain x n (t): where x n (t) is the normalized signal, and the normalized signal .

[0008] Preferably, in step 2, a center-symmetric, equidistant dither sequence with a length of N d is constructed: where LSB is the quantization step of the uniform quantizer, and the average value of the dither sequence is 0.

[0009] Preferably, in step 3, for each dither sequence u i in the dither sequence constructed in step 2, a perturbed signal is constructed: and then uniformly quantized to obtain : N d groups of quantization results are obtained. where x i (t) represents the signal after perturbation, is the quantization function of the uniform quantizer.

[0010] Preferably, in step 4, all the quantization sampling results obtained in step 3 are averaged by sample dimension: to obtain the quantization signal.

[0011] Preferably, in step 5, the signal obtained in step 4 is subjected to inverse transform processing: to restore the signal to the original amplitude space.

[0012] Preferably, step 6 is performed after step 5: the signal obtained in step 5 is input into a subsequent communication system model (including demodulation, bit error statistics and spectrum analysis, etc.) to evaluate the performance of the method of the present invention in a real optical transmission system.

[0013] This invention also discloses a digital super-resolution low-bit quantization system based on multi-offset observation fusion for performing the above method, which includes the following modules: Amplitude normalization module: Performs amplitude normalization processing on the signal to be quantized based on the amplitude range of the signal, the input dynamic range of the uniform quantizer, and the quantization step size LSB. Jitter Sequence Module: Constructs a centrally symmetric, equally spaced jitter sequence based on the quantization step size LSB, with its value range covering [-LSB / 2, LSB / 2]; Uniform quantization module: Superimposes each jitter sequence in the jitter sequence module onto the normalized signal to form several groups of perturbed signals, and performs uniform quantization on each group of signals under the same quantization threshold structure; Output signal sequence construction module: Averages multiple quantization results obtained from the uniform quantization module in the sample dimension to construct an output signal sequence with reduced equivalent quantization error variance; Inverse Transform Module: Performs inverse transform processing on the output signal obtained from the output signal sequence construction module, so that it returns to the original amplitude space.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. Significantly improved quantization performance: This invention constructs a centrally symmetric, equally spaced jitter sequence to independently quantize and average the signal multiple times within a quantization step, which can significantly reduce the quantization error variance and homogenize the noise distribution. Without changing the quantizer structure, it effectively improves the quantization accuracy of low-bit broadband signals, and is particularly suitable for low-bit quantization scenarios of broadband multi-carrier signals such as OFDM.

[0015] 2. Low hardware implementation complexity: The jitter sequence of this invention has a regular linear structure and can be directly obtained through a simple adder or lookup table, without the need for additional complex transformations, circuits, or high-order noise shaping modules. The quantization process can be achieved by repeatedly using the same quantizer and performing accumulation and shift averaging at the output. The hardware logic is simple, resource consumption is low, and it is easy to deploy quickly on real-time systems.

[0016] 3. Strong structural versatility and scalability: This invention does not rely on specific quantizer structures, bit width settings, or signal formats, and can be directly embedded into existing communication and signal processing links. By adjusting the jitter sequence length, a flexible trade-off can be achieved between performance improvement and hardware overhead, making this invention exhibit good portability and scalability in various low-bit quantization systems such as broadband communication, millimeter-wave radar, and edge intelligence. Attached Figure Description

[0017] Figure 1 This is a flowchart of a preferred embodiment of the present invention, which describes a digital super-resolution low-bit quantization method based on multi-offset observation fusion.

[0018] Figure 2 This is a comparison chart of the bit error rate of a real-time system at 3-8 bit resolution when the present invention is applied to traditional uniform quantization and non-uniform quantization algorithms.

[0019] Figure 3 This is a block diagram of a digital super-resolution low-bit quantization system based on multi-offset observation fusion, according to a preferred embodiment of the present invention. Detailed Implementation

[0020] To more clearly illustrate the embodiments of the present invention, specific implementation methods will be described below with reference to the accompanying drawings. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without any creative effort.

[0021] like Figure 1 As shown in the figure, this embodiment proposes a digital super-resolution low-bit quantization method based on multi-offset observation fusion, the specific steps of which are as follows: Step 1: Based on the amplitude range of the time-domain signal x(t) and the quantization range of the uniform quantizer, perform a uniform normalization process on the signal to be quantized x(t) to obtain x. n (t). By performing amplitude normalization on the OFDM signal, the overall dynamic range of the signal is stabilized within a fixed range, which facilitates the construction of a unified quantization range and threshold structure: Where x(t) is the time-domain OFDM signal output by the OFDM transmitter, x n (t) represents the normalized signal. ; Step 2: Construct a quantization step size (LSB) of length N. d A specific jitter sequence that is centrally symmetric and equally spaced: Its range covers [-LSB / 2, LSB / 2], ensuring effective elimination of overall quantization error. Here, LSB is the quantization step size of the uniform quantizer, and the average value of the jitter sequence is 0, i.e. ; Step 3: For each perturbation sequence value u constructed in Step 2 i The signal after the perturbation is constructed: And perform uniform quantization using the same threshold structure: get , to obtain N d Group Independent Quantization Output Among them, x i (t) represents the scrambled signal. It is the quantization function of the uniform quantizer.

[0022] Step 4: Average all the quantization sampling results obtained in Step 3 according to the sample dimensions to obtain: An output signal with reduced equivalent error variance is obtained. The quantized signal obtained above has a significantly reduced mean quantization error, is closer to a white noise distribution, and has greatly reduced signal distortion.

[0023] Step 5: Perform an inverse transform on the scrambled and averaged quantized signal obtained in Step 4: Restore the signal to its original amplitude space.

[0024] The improved quantized signal is sent to an OFDM receiver to perform demodulation, FFT and reception processes. By comparing it with the quantized signal without jitter, the performance improvement effect of this method in terms of bit error rate is verified.

[0025] Based on the above technical solution, this invention verifies the feasibility and superiority of a digital super-resolution low-bit quantization scheme based on multi-offset observation fusion using a real-time IMDD-OFDM system. Figure 2 The figure shows a comparison of the bit error rate (BER) of the proposed solution at 3-8 bit resolution in real-time systems when applied to traditional uniform quantization and non-uniform quantization algorithms. As can be seen from the figure, the proposed solution, combining uniform and non-uniform quantization, achieves BER below the forward error correction (FEC) threshold at 3-8 bit resolution, ensuring error-free transmission in the communication system through FEC. The BER curve of the proposed solution at 3 bit resolution is close to that of traditional methods at high resolution (5-8 bits), demonstrating that the proposed solution can improve the accuracy of traditional quantization methods by at least 2 bits with low implementation complexity, significantly improving DAC / ADC quantization performance and reducing the deterioration of real-time system performance caused by quantization noise.

[0026] like Figure 3 As shown, this embodiment proposes a digital super-resolution low-bit quantization system based on multi-offset observation fusion to execute the above method embodiment. The system includes the following modules: Amplitude normalization module: Performs amplitude normalization processing on the signal to be quantized based on the amplitude range of the signal, the input dynamic range of the uniform quantizer, and the quantization step size LSB. Jitter Sequence Module: Constructs a centrally symmetric, equally spaced jitter sequence based on the quantization step size LSB, with its value range covering [-LSB / 2, LSB / 2]; Uniform quantization module: Superimposes each jitter sequence in the jitter sequence module onto the normalized signal to form several groups of perturbed signals, and performs uniform quantization on each group of signals under the same quantization threshold structure; Output signal sequence construction module: Averages multiple quantization results obtained from the uniform quantization module in the sample dimension to construct an output signal sequence with reduced equivalent quantization error variance; Inverse Transform Module: Performs inverse transform processing on the output signal obtained from the output signal sequence construction module, so that it returns to the original amplitude space.

[0027] Other aspects of this embodiment can be found in the above method embodiments.

[0028] In summary, the technical solution of this invention first establishes a unified normalization process for the signal to be quantized based on the signal amplitude range, constructs a specific jitter sequence with central symmetry and equal spacing based on the quantization step size, adds jitter to each signal for each jitter value, performs independent quantization and averages it according to the sample dimension, and finally performs inverse transformation and subsequent recovery operations to construct a DAC design with low bit precision, low cost and high fidelity.

[0029] It should be noted that the above description is only a detailed explanation of the preferred embodiments and principles of the present invention. For those skilled in the art, there may be changes in the specific implementation based on the ideas provided by the present invention, and these changes should also be considered within the scope of protection of the present invention.

Claims

1. A digital super-resolution low-bit quantization method based on multi-misalignment observation fusion, characterized by: Follow these steps: Step 1: Perform amplitude normalization on the signal to be quantized based on the signal amplitude range, the input dynamic range of the uniform quantizer, and the quantization step size LSB. Step 2: Construct a centrally symmetric, equally spaced jitter sequence based on the quantization step size LSB, with its value range covering [-LSB / 2, LSB / 2]; Step 3: Superimpose each jitter sequence from Step 2 onto the normalized signal to form several groups of perturbed signals, and perform uniform quantization on each group of signals under the same quantization threshold structure; Step 4: Average the multiple quantization results obtained in Step 3 along the sample dimension to construct an output signal sequence with reduced equivalent quantization error variance; Step 5: Perform an inverse transform on the output signal obtained in Step 4 to bring it back to the original amplitude space.

2. The digital super-resolution low-bit quantization method based on multi-offset observation fusion as described in claim 1, characterized in that, In step 1, based on the amplitude range of the time-domain signal x(t) and the quantization step size LSB of the uniform quantizer, the amplitude of the signal to be quantized x(t) is normalized to obtain x. n (t): Where, x n (t) represents the normalized signal. .

3. The digital super-resolution low-bit quantization method based on multi-offset observation fusion as described in claim 2, characterized in that, In step 2, the construction length is N. d Centrally symmetric, equally spaced jitter sequences: Here, LSB is the quantization step size of the uniform quantizer, and the average value of the jitter sequence is 0.

4. The digital super-resolution low-bit quantization method based on multi-offset observation fusion as described in claim 3, characterized in that, In step 3, for each jitter sequence u in the jitter sequence constructed in step 2 i Construct the signal after perturbation; Then perform uniform quantization to obtain : Get N d Group quantification results ; Where, x i (t) represents the scrambled signal. It is the quantization function of the uniform quantizer.

5. The digital super-resolution low-bit quantization method based on multi-offset observation fusion as described in claim 4, characterized in that, In step 4, all the quantization sampling results obtained in step 3 are averaged according to the sample dimensions: The obtained quantized signal.

6. The digital super-resolution low-bit quantization method based on multi-offset observation fusion as described in claim 5, characterized in that, In step 5, the signal obtained in step 4 is subjected to inverse transformation processing; Restore the signal to its original amplitude space.

7. A digital super-resolution low-bit quantization system based on multi-offset observation fusion, used to perform the method as described in any one of claims 1-6, characterized in that, Includes the following modules: Amplitude normalization module: Performs amplitude normalization processing on the signal to be quantized based on the amplitude range of the signal, the input dynamic range of the uniform quantizer, and the quantization step size LSB. Jitter Sequence Module: Constructs a centrally symmetric, equally spaced jitter sequence based on the quantization step size LSB, with its value range covering [-LSB / 2, LSB / 2]; Uniform quantization module: Superimposes each jitter sequence in the jitter sequence module onto the normalized signal to form several groups of perturbed signals, and performs uniform quantization on each group of signals under the same quantization threshold structure; Output signal sequence construction module: Averages multiple quantization results obtained from the uniform quantization module in the sample dimension to construct an output signal sequence with reduced equivalent quantization error variance; Inverse Transform Module: Performs inverse transform processing on the output signal obtained from the output signal sequence construction module, so that it returns to the original amplitude space.

Images