Solve Mismatch Problem in Compressed Sensing

Le Yang

2024-10-15

Abstract:This article proposes a novel algorithm for solving mismatch problem in compressed sensing. Its core is to transform mismatch problem into matched by constructing a new measurement matrix to match measurement value under unknown measurement matrix. Therefore, we propose mismatch equation and establish two types of algorithm based on it, which are matched solution of unknown measurement matrix and calibration of unknown measurement matrix. Experiments have shown that when under low gaussian noise levels, the constructed measurement matrix can transform the mismatch problem into matched and recover original images. The code is available: <a class="link-external link-https" href="https://github.com/yanglebupt/mismatch-solution" rel="external noopener nofollow">this https URL</a>

Signal Processing,Information Theory

What problem does this paper attempt to address?

This paper attempts to solve the Mismatch Problem in compressed sensing. Specifically, when the measurement matrix \( A \) and the measurement value \( y \) do not correspond, the traditional compressed sensing algorithm cannot accurately recover the original signal \( x \). This mismatch problem is very common in practical applications, especially when the measurement matrix is unknown.

### Main contributions of the paper

1. **Propose a new algorithm framework**: The paper proposes a method to transform the mismatch problem into a matching problem by constructing a new measurement matrix \( A_{\text{recv}} \). The core idea is to estimate the unknown measurement matrix through the known measurement values and measurement matrix, so that the mismatch problem can be transformed into a matching problem within a certain error range.

2. **Two types of solutions**:
- **Matching solution for unknown measurement matrix**: By constructing a new measurement matrix \( A_{\text{recv}} \), the measurement value \( y \) and the new matrix \( A_{\text{recv}} \) form a matching pair, so that the traditional compressed sensing algorithm can be used to recover the original image.
- **Calibration method for unknown measurement matrix**: By calibrating the unknown measurement matrix \( A_u \), the calibrated matrix \( A_{\text{recv}} \) can be applied to any unknown image, thereby achieving accurate recovery of measurement values.

3. **Theoretical analysis and experimental verification**: The paper not only provides detailed theoretical derivations but also verifies the effectiveness of the proposed method through experiments. The experimental results show that under a low - Gaussian noise level, the constructed measurement matrix \( A_{\text{recv}} \) can successfully recover the original image.

### Key formulas

- **Basic formula of compressed sensing**:
\[
y = Ax+\epsilon
\]
where \( A\in\mathbb{R}^{M\times N} \) is the measurement matrix, \( x\in\mathbb{R}^{N\times 1} \) is the unknown image, \( y\in\mathbb{R}^{M\times 1} \) is the measurement value, and \( \epsilon\in\mathbb{R}^{M\times 1} \) is the system noise.

- **Mismatch equation**:
\[
A_{\text{recv}}=\frac{y_0^T\Sigma y_0}{y_0^T\Sigma A}
\]
where \( \Sigma\in\mathbb{R}^{M\times M} \) is any non - zero matrix, and \( y_0 = A\cdot\text{PMimage} \) is the pre - measurement value.

- **Update formula in the iterative algorithm**:
\[
A_{\text{recv}}^{(k + 1)}=A_{\text{recv}}^{(k)}+\frac{y_0^T\Sigma e_y^{(k)}}{y_0^T\Sigma y_0}y_0^T\Sigma A
\]
where \( e_y^{(k)}=y' - A_{\text{recv}}^{(k)}x' \) is the measurement error at the \( k \) - th iteration.

### Experimental results

- **Recovery effect in the absence of noise**: In the absence of noise, the original image can be successfully recovered using the constructed measurement matrix \( A_{\text{recv}} \).
- **Influence of noise**: Under a low - noise level, the constructed measurement matrix can still recover the image relatively well; however, under a high - noise level, the recovery effect will decline significantly.
- **Influence of device precision**: Devices with different precisions (such as RTX 2080 Ti, RTXA400, RTX3090) have a significant impact on the performance of the algorithm. On medium - precision devices, better results can be obtained by using float32 for calculation.