Method of non-uniform wavelet bandpass sampling

Patent metadata

Field	Value
Publication number	US-10020930-B2
Application number	US-201615343613-A
Country	US
Kind code	B2
Filing date	Nov 4, 2016
Priority date	Nov 4, 2016
Publication date	Jul 10, 2018
Grant date	Jul 10, 2018

Abstract

A compressed sensing method based on non-uniform wavelet bandpass sampling. A K-sparse signal of interest is projected onto a sequence of waveforms succeeding one another at the bandpass sampling rate, the waveforms belonging to an overcomplete dictionary, the parameters of the waveforms depending on the characteristics of the bands of the signal. The correlation values are then non-uniformly sampled to provide a compressed representation of the signal.

First claim

The invention claimed is:

1. A compressed sensing method for compressed sensing a K-sparse signal, the spectrum of which comprising a plurality (N b ) of bands, said method comprising:
projecting the signal onto a plurality N Ω of waveforms succeeding one another at a bandpass sampling rate of the signal f Ω =λ(Ω) where Ω is the spectral support of the signal and λ(.) is the Lebesgue measure, to produce a plurality N Ω of projection values, the waveforms belonging to a dictionary, the parameters of the waveforms depending upon the characteristics of the bands of the signal;
sampling the projection values non-uniformly in time, M<N Ω projection values being selected at random among N Ω projection values, to provide a compressed representation of the signal.

2. The compressed sensing method according to claim 1 , wherein the pulse width of the waveforms is chosen shorter than the inverse of an aggregated bandwidth of the signal, the aggregated bandwidth being defined as the sum of the bandwidths of the respective bands of the signal.

3. The compressed sensing method according to claim 1 , wherein the N Ω waveforms are selected from the waveforms of the dictionary such that the mutual coherence between sampling matrix and inverse Fourier transform matrix restricted to the spectral support of the signal is minimal.

4. The compressed sensing method according to claim 3 , wherein the N Ω waveforms are such that the ratio between the waveform bandwidth and an aggregated bandwidth of the signal is higher than 50%, the aggregated bandwidth being defined as the sum of the bandwidths of the bands of the signal.

5. The compressed sensing method according to claim 3 , wherein the N Ω waveforms are distributed among a plurality of sets, each set being associated with a band of the plurality of bands.

6. The compressed sensing method according to claim 5 , wherein the N Ω waveforms form a sequence at the bandpass sampling rate, the waveforms belonging to a set associated with a band B i of the signal recurring at a period

B
agg
w

B
i
w

⁢

T
Ω

where B i w is the width of band B i , B agg w is an aggregated bandwidth, defined as the sum of the bandwidths of the bands of the signal, and T Ω is the inverse of the bandpass sampling rate.

7. The compressed sensing method according to claim 5 , wherein the N Ω waveforms are distributed over a plurality P of subsequences, each subsequence being formed by waveforms succeeding one another at one P th of the bandpass sampling rate, the signal being projected in P parallel branches on said plurality P of subsequences.

8. The compressed sensing method according to claim 5 , wherein the N Ω waveforms are distributed over a plurality P of subsequences, each subsequence being formed by waveforms having the same parameters and succeeding one another at a rate higher or equal to one P th of the bandpass sampling rate, the signal being projected in P parallel branches on said plurality P of subsequences.

9. The compressed sensing method according to claim 1 , wherein the waveforms are Gabor waveforms.

10. The compressed sensing method according to claim 1 , wherein the waveforms are Morlet or C-Morlet wavelets.

11. The compressed sensing method according to claim 1 , wherein the waveforms are Haar wavelets.

12. The compressed sensing method according to claim 1 , wherein the selection of a projection value at a given time is performed according to a logical value output by a pseudo-random generator at that time.

13. The compressed sensing method according to claim 1 , wherein, prior to projecting the signal onto the plurality of waveforms, K non-zero components of the sparse signal are searched by an iterative process, starting with a plurality of frequency bins covering the signal spectrum, each iteration step comprising a division of each frequency bin into a first frequency half-bin of lower frequency and a second frequency half-bin of higher frequency, a first wavelet being associated with the first half-bin and a second wavelet being associated with the second half-bin, the signal being then projected onto the first and the second wavelets to obtain a first projection value and a second projection value, the first and second values being compared to each other, and only the half-bin corresponding to the higher value being retained for the next iteration step.

14. The compressed sensing method according to claim 13 , wherein, at each iteration step, the first wavelet is centered on the center of the first half-bin and the second wavelet is centered on the center of the second half-bin.

15. The compressed sensing method according to claim 14 , wherein the iteration process is carried out until the size of the frequency bins reaches a predetermined value, the support of the spectral spectrum being formed by the frequency bins retained at the last iteration step, the bands of the signal spectrum being derived from this support.

16. The compressed sensing method according to claim 13 , wherein, the signal being real, the first wavelet is centered on a first frequency shifted from the center of the first half-bin by a predetermined frequency shift, and the second wavelet is centered on a second frequency shifted from the center of the second half-bin by said predetermined frequency shift, and if the first and second values in a frequency bin are close to each other within a given margin, the corresponding first and second values obtained for the symmetric bin are also compared to determine whether the first-half or second-half bin is retained.

Assignees

• Commissariat Energie Atomique
• Univ Cornell

Inventors

• Pelissier Michael
• Studer Christoph Emanuel

Classifications

• H04L27/12

  Modulator circuits; Transmitter circuits · CPC title

• H03M7/3062

  Compressive sampling or sensing · CPC title

• H04L12/12

  Arrangements for remote connection or disconnection of substations or of equipment thereof · CPC title

• H04L27/0004

  using wavelets · CPC title

• H04L7/0016 Primary

  correction of synchronization errors · CPC title