A Deep Learning Framework of Quantized Compressed Sensing for Wireless Neural Recording

Sun, Biao; Feng, Hui; Chen, Kefan; Zhu, Xiaoyan

doi:10.1109/access.2016.2604397

Cited by 64 publications

(55 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In particular, [15]- [18] study linear compression for real signals; [19]- [21] consider nonlinear compression for real signals [19], [20] and complex signals [21]. Note that existing joint signal compression and recovery methods [15]- [21] cannot provide linear compression for complex signals, and the extensions to joint linear compression and recovery methods for complex signals estimation are not trivial. In addition, the optimal measurement matrix and recovery method for sparse signal estimation are not necessarily always the best for support recovery.…”

Section: Introductionmentioning

confidence: 99%

Joint Design of Measurement Matrix and Sparse Support Recovery Method via Deep Auto-Encoder

Zhang

Cui

et al. 2019

IEEE Signal Process. Lett.

View full text Add to dashboard Cite

Sparse support recovery arises in many applications in communications and signal processing. Existing methods tackle sparse support recovery problems for a given measurement matrix, and cannot flexibly exploit the properties of sparsity patterns for improving performance. In this letter, we propose a datadriven approach to jointly design the measurement matrix and support recovery method for complex sparse signals, using autoencoder in deep learning. The proposed architecture includes two components, an auto-encoder and a hard thresholding module. The proposed auto-encoder successfully handles complex signals using standard auto-encoder for real numbers. The proposed approach can effectively exploit properties of sparsity patterns, and is especially useful when these underlying properties do not have analytic models. In addition, the proposed approach can achieve sparse support recovery with low computational complexity. Experiments are conducted on an application example, device activity detection in grant-free massive access for massive machine type communications (mMTC). Numerical results show that the proposed approach achieves significantly better performance with much less computation time than classic methods, in the presence of extra structures in sparsity patterns.

show abstract

Section: Introductionmentioning

confidence: 99%

Joint Design of Measurement Matrix and Sparse Support Recovery Method via Deep Auto-Encoder

Zhang

Cui

et al. 2019

IEEE Signal Process. Lett.

View full text Add to dashboard Cite

show abstract

“…Here, interpolation or approximation methods can be used. In this work, we focus on parameterization methods for global interpolation and attempt to improve the centripetal (exponential) method by converting it from a static exponential form to a dynamic exponential form using some chord properties [1]- [5].…”

Section: Introductionmentioning

confidence: 99%

Dynamic Centripetal Parameterization Method for B-Spline Curve Interpolation

et al. 2020

View full text Add to dashboard Cite

B-spline data interpolation and approximation require parameterization at the first step. For this purpose, many algorithms have been developed, such as the uniform, centripetal, chord length, Foley and universal methods. The uniform method works well if the input data points are distributed regularly. The chord length method produces large deflections if long chords exist in the data polygon. To remove this effect, the centripetal method was developed. The traditional centripetal method uses a fixed power for chord lengths for parameter distribution. In this paper, we propose an improved version of the centripetal parameterization method for B-spline data interpolation. Our experiments show that individual dynamic power calculation can be possible for each chord length. This new parameterization method produces better behavior when compared to the traditional centripetal method and is more robust against fast changes in chord lengths since it uses the natural logarithm of chord lengths to calculate the parameters. INDEX TERMS B-spline curves, parameterization, interpolation.

show abstract

“…Despite remarkable advances in non-quantized CS, only a few works have applied deep learning for QCS. The first end-to-end QCS design was proposed in [24], where the devised method optimizes a binary measurement acquisition realized by a DNN, a compander-based non-uniform quantizer, and a DNN decoder to estimate neural spikes.…”

Section: Introductionmentioning

confidence: 99%

Low-Complexity Vector Quantized Compressed Sensing via Deep Neural Networks

Leinonen¹,

Codreanu²

2020

IEEE Open J. Commun. Soc.

View full text Add to dashboard Cite

Sparse signals, encountered in many wireless and signal acquisition applications, can be acquired via compressed sensing (CS) to reduce computations and transmissions, crucial for resource-limited devices, e.g., wireless sensors. Since the information signals are often continuous-valued, digital communication of compressive measurements requires quantization. In such a quantized compressed sensing (QCS) context, we address remote acquisition of a sparse source through vector quantized noisy compressive measurements. We propose a deep encoder-decoder architecture, consisting of an encoder deep neural network (DNN), a quantizer, and a decoder DNN, that realizes low-complexity vector quantization aiming at minimizing the mean-square error of the signal reconstruction for a given quantization rate. We devise a supervised learning method using stochastic gradient descent and backpropagation to train the system blocks. Strategies to overcome the vanishing gradient problem are proposed. Simulation results show that the proposed non-iterative DNN-based QCS method achieves higher rate-distortion performance with lower algorithm complexity as compared to standard QCS methods, conducive to delay-sensitive applications with large-scale signals.

show abstract

A Deep Learning Framework of Quantized Compressed Sensing for Wireless Neural Recording

Cited by 64 publications

References 42 publications

Joint Design of Measurement Matrix and Sparse Support Recovery Method via Deep Auto-Encoder

Joint Design of Measurement Matrix and Sparse Support Recovery Method via Deep Auto-Encoder

Dynamic Centripetal Parameterization Method for B-Spline Curve Interpolation

Low-Complexity Vector Quantized Compressed Sensing via Deep Neural Networks

Contact Info

Product

Resources

About