Multi-channel EEG compression based on 3D decompositions
Summary (1 min read)
Introduction
- Various compression algorithms for multi-channel electroencephalograms (EEG) are proposed and compared.
- A general two-stage coding framework is developed for multi-channel EEG compression.
- In Section 2 the authors explain the multi-way (or volumetric data) representation of multi-channel EEG.
- The authors stack the matrices from subsequent time instances to form a volume, as shown in Fig. 1(b).
- In the following the authors explain their three compression algorithms.
3.1.1. Volumetric Coding Approach
- Fig. 2 shows a diagram of the proposed two-stage coder for multi-channel EEG signals.
- The compressed dataIen is then decoded, yielding the reconstructed dataIl.
- The residual ε is uniformly quantized to generate quantization indicesεq, with maximum error no larger thanδ: Table 1.
- Ni is the number of elements inth subband (d) Coding orderO = descend(RED) (e) Code the subband according to O by Arithmetic coding (b) RE= RE+ REDO(i) ·NO(i) // update relative energy of the coded subband.
- The pre-processing step,i.e., formation of tensor from multi-channel EEG, is the principle difference from the coders used in image compression [7].
3.1.2. Subband Specific Arithmetic Coding (SAC)
- The authors first order the wavelet subbands based on their relative energy density (RED).
- The remaining subbands are less significant in terms of their RED; the authors first quantize them (cf. (5)), and then apply arithmetic coding.
- This two-stage procedure results in lossy compression of the EEG signals.
- It is noteworthy that in the second coding step, the authors quantize the wavelet subbands; this may lead to a substantial error in time domain.
- In other words, the authors cannot control the maximum distortion in time domain through this approach.
3.2. Tensor-based Compression
- The authors apply parallel factor decomposition decomposition [8] to the three-way tensorI, formed from multichannel EEG.
- The PARAFAC based decomposition of a three-way tensor is given by: I = r ∑ i=1 ai ◦ bi ◦ ci + E , (7) whereE represents the residual tensor, and, b, andc represent the factors along the three modes, whereas◦ stands for the outer-product along the particular mode.
- The residual quantization is performed in time-domain for volumetric and PARAFAC coding, whereas in SAC, quantization is performed in wavelet domain.
- The authors have presented novel compression schemes for multichannel EEG.
Did you find this useful? Give us your feedback
Citations
21 citations
Cites methods from "Multi-channel EEG compression based..."
...in [18] utilises wavelet-based volumetric coding, energy-based lossless compression of wavelet subbands and tensor decomposition based coding for effective compression....
[...]
8 citations
Cites methods from "Multi-channel EEG compression based..."
...The proposed technique drastically reduces the amount of sampled data at each node, thus allowing the nodes to spend more time in a low-power sleep mode and save energy....
[...]
2 citations
References
11,407 citations
9,227 citations
"Multi-channel EEG compression based..." refers methods in this paper
...We apply parallel factor decomposition (PARAFAC) decomposition [8] to the three-way tensor I, formed from multichannel EEG....
[...]
3,211 citations
"Multi-channel EEG compression based..." refers methods in this paper
...In the first stage, we consider (i) wavelet-based volumetric coding; (ii) energybased lossless compression of wavelet subbands; (iii) tensor decomposition based coding....
[...]
3,195 citations
178 citations
Related Papers (5)
Frequently Asked Questions (10)
Q2. What is the main idea of the paper?
The main idea is to exploit the intra- and inter-channel correlations simultaneously by arranging the multi-channel EEG as a volume, and to represent that volume in different ways.
Q3. What is the way to encode a wavelet subband?
The tensor-based coding scheme yields smaller worst-case error than both subband specific coding and volumetric coding, yet the average error is only slightly larger than in subband specific coding and much smaller than in volumetric coding.
Q4. How many samples are used in the algorithm?
(10)The authors consider segments of 1024 samples from each channel, arranged in a suitable volume size, specifically, 32×32× 64 for t/dt/s volume, 8 × 8 × 1024 for s/s/t volume.
Q5. What is the coding algorithm used for the wavelet?
The authors consider three lossy compression algorithms (Stage 1): (i) 3D Wavelet volumetric coding, (ii) 3D Wavelet subband specific arithmetic coding, and (iii) tensor decomposition (PARAFAC) based coding.
Q6. What are the two specific ways to extract a volumetric data from multi-channel EEG?
The authors consider two specific ways to extract a volumetric data from multi-channel EEG, where the three axes capture spatial and temporal variations in different form.
Q7. How is the quality of the reconstructed signal assessed?
The quality of the reconstructed signal (x̃) is assessed using percent rootmean-square distortion (PRD (%)):PRD (%) =√ √ √ √ ∑N i=1(x(i)− x̃(i)) 2∑N i=1 x(i)2 × 100. (9)The authors also use an alternative quantitative distortion measure, based on the maximum absolute difference between x and x̃:PSNR(x, x̃) = 10 log10(2Q − 1max(|x− x̃|)).
Q8. What is the main challenge for multi-channel EEG?
In their previous work [2], the authors introduced a pre-processing technique where single-channel EEG is arranged as a matrix before compression; this representation improved the RateDistortion (R-D) performance over conventional compression schemes.
Q9. What is the energy threshold for the subband specific coding?
The energy threshold (τ ) for the subband specific arithmetic coding is fixed to 50%; the authors obtained the best results for that value of the threshold.
Q10. What is the arithmetic coding scheme used in the algorithm?
The authors analyze the performance of the algorithms based on compression ratio:CR = LorigLcomp , (8)where Lorig and Lcomp are the bit length of original and reconstructed multi-channel EEG signals respectively.