๋ณธ๋ฌธ ๋ฐ”๋กœ๊ฐ€๊ธฐ

๐Ÿ”ฌ Science/๐Ÿ“ป Signal

์ƒ์ฒด ์‹ ํ˜ธ ๋ ˆ์ด๋” ์‹œ์Šคํ…œ์„ ์‚ฌ์šฉํ•œ ์ˆ˜๋ฉด ๋ถ„๋ฅ˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์ ์šฉ

์ฐธ๊ณ  ๋…ผ๋ฌธ : R.de Goederen et al(2021), "Radar-based sleep stage classification in children undergoing polysomnography: a pilot-study", Sleep Medicine 82

 

XeThru X2M200 & X4M200 ๋ ˆ์ด๋” ๋ชจ๋“ˆ

-> ์‚ฌ๋žŒ์˜ resting and breathing ์ƒํƒœ์ผ ๋•Œ์˜ ์ฃผ๊ธฐ์ ์ธ ์›€์ง์ž„์„ ๊ด€์ฐฐํ•จ

์ด ๋ชจ๋“ˆ์€ local oscillator์—์„œ ์ƒ์„ฑ๋œ ์ผ์ •ํ•œ ํŽ„์Šค๊ฐ€ ์•ˆํ…Œ๋‚˜๋ฅผ ํ†ตํ•ด ์ „์†ก๋˜๋Š” pulse-Doppler processing์„ ์‚ฌ์šฉํ•จ

 

ํ•ด๋‹น ํŽ„์Šค๋Š” ๋ฐ˜์‚ฌํŒ์„ ๋งŒ๋‚  ๋•Œ๊นŒ์ง€ ๊ณต๊ฐ„์„ ํ†ตํ•ด ์ „ํŒŒ๋˜๋Š”๋ฐ,

์ „์†ก๋œ ์—๋„ˆ์ง€ ์ค‘ ์ผ๋ถ€๋Š” ์›€์ง์ž„์œผ๋กœ ์ธํ•œ ์œ„์ƒ ๋ณ€์กฐ์™€ ํ•จ๊ป˜ ์ˆ˜์‹ ๊ธฐ๋กœ ๋‹ค์‹œ ๋ฐ˜์‚ฌ๋œ๋‹ค. 

์ˆ˜์‹ ๋œ RF -> local oscillator์— ์˜ํ•ด baseband signal๋กœ down-converted.

baseband signal์€ ๋‘ ๊ฐœ์˜ ์ง๊ต ์‹ ํ˜ธ๋กœ ๋‚˜๋ˆ ์ง.

20Hz in X2M200, 17Hz in X4M200

๋‘ ๋ ˆ์ด๋” ๋ชจ๋‘ In-Phase and Quadrature (IQ) or Amplitude/phase (AP) data ์ œ๊ณต

=> ์ด์ค‘ Amplitude ์‚ฌ์šฉํ•จ. 

 

Body motion๊ณผ respiration rate์€ ์œ„์˜ baseband data๋ฅผ processingํ•ด์„œ ์–ป์–ด์ง.

Motion signal <- ์ง„ํญ baseband signal์˜ ๋‘ ๊ฐœ์˜ ํ›„์† ์‹œ๊ฐ„ frame ๊ฐ„ ์ฐจ์ด๋ฅผ ํ•ฉ์ณ์„œ ์ƒ์„ฑํ•จ.

์‹์€ ์•„๋ž˜์™€ ๊ฐ™์ด ํ‘œํ˜„. 

$$ {MVM}win = \sum_{t = t{i}-win+1 }^{ti}|[A(t) - A(t-1)]| $$ 

(MVM : ํŠน์ • ์ƒ˜ํ”Œ ํฌ์ธํŠธ t์—์„œ time window (win)์•ˆ์˜ movement quantity ํ‘œํ˜„;

A : baseband data์˜ amplitude;

ti : ์‹œ๊ฐ„ ๋‚ด ์‹œ๊ฐ„ ํฌ์ธํŠธ ์ง€์ •;)

 

๊ทธ ๋‹ค์Œ, 'range-frequency' matrix๋ฅผ ๋งŒ๋“ค๊ธฐ ์œ„ํ•ด

baseband amplitude ๋ฐ์ดํ„ฐ์— ๋Œ€ํ•œ A Fast Fourier Transform (FFT) ๋ถ„์„๋ฒ•์„ ์ˆ˜ํ–‰ํ•œ๋‹ค.

์ •์ ์ธ object๋Š” ์ง€์›Œ์ง€๊ณ , ํ˜ธํก์˜ ์ž‘์€ ์›€์ง์ž„์€ ์ธก์ •๋œ๋‹ค.

range-frequency matrix์˜ ๋กœ์ปฌ๋ผ์ด์ง• ๋ฐฉ๋ฒ•์„ ํ†ตํ•ด BR(breathing rate in respirations / minute)์ด ์ถ”์ ๋œ๋‹ค.

 

+ Feature Extraction

38๊ฐœ์˜ ๋™์ž‘ ๋ฐ ํ˜ธํก ํŠน์ง•์ด ์ถ”์ถœ๋จ. 9๊ฐœ์˜ ํŠน์ง• from ๋™์ž‘ ์‹ ํ˜ธ / 29๊ฐœ์˜ ํŠน์ง• from BR

 

1) Respiration Feature : 6๋ถ„ ๊ฐ„๊ฒฉ์˜ BR ๋ถ„์‚ฐ๊ณผ ํ‰๊ท  ๊ณ„์‚ฐ. ํ•œ epoch์— ๊ฑธ์ณ ๊ณ„์‚ฐ๋จ.

์‹œ๊ฐ„ ์ง€์—ฐ์ด ๋‹ค๋ฅธ 1~10๊นŒ์ง€ scaling๋œ ์ƒ˜ํ”Œ entropy, ๊ทผ์‚ฌ entropy, multiscale ์ƒ˜ํ”Œ entropy๊ฐ€ ํ•œ epoch์— ๊ฑธ์ณ ๊ณ„์‚ฐ๋จ.

- ์ด ํŠน์ง•๋“ค์€ time-series signal์˜ ๋ณต์žก์„ฑ๊ณผ ์œ ์‚ฌ์„ฑ์„ ์ธก์ •ํ•˜๊ธฐ ์œ„ํ•ด ๋‹ค์–‘ํ•œ ๋ฐฉ๋ฒ•์œผ๋กœ ์–ป์Œ. 

- ํ•œ epoch ๋‚ด์˜ time domain๊ณผ frequency domain ๋ชจ๋‘์—์„œ Teager energy์˜ ํ‰๊ท ๊ฐ’์ด ๊ณ„์‚ฐ๋จ.

- signal์˜ fractal dimension์„ ์ถ”์ •ํ•˜๊ธฐ ์œ„ํ•ด Katz's fractal dimension ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์‚ฌ์šฉ

-> ๋ชจ๋“  feature๋Š” ๊ฐ recording์—์„œ feature์˜ ํ‰๊ท ๊ฐ’์„ ๋นผ๊ณ , ํ‘œ์ค€ํŽธ์ฐจ๋กœ ๋‚˜๋ˆ„๋Š” Z-score normalization์„ ๊ฑฐ์นจ.

(์ •๊ทœํ™” ๋‹จ๊ณ„๋Š” ๋ฐ์ดํ„ฐ์˜ ์ค‘๋ณต์„ ์ค„์ด๊ณ , ๋ฌด๊ฒฐ์„ฑ์„ ๊ฐœ์„ ํ•ด ํ”ผํ—˜์ž ๊ฐ„ ์ƒ๋ฆฌ์  ํŽธ์ฐจ๋ฅผ ์ค„์ด๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•จ)

 

2) Motion Feature : baseband ๋ถ„์„ ๊ฒฐ๊ณผ, motion detection ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ์„œ๋กœ ๋‹ค๋ฅธ ์„ค์ •์—์„œ ๋‘ ๊ฐ€์ง€ motion signal์„ ์–ป์Œ.

a. ํ•˜๋‚˜๋Š” window๊ฐ€ 1์ดˆ์ธ ์‹ ํ˜ธ(1์ดˆ ์ด๋‚ด์˜ ์›€์ง์ž„ ์–‘์„ ๋‚˜ํƒ€๋‚ด๋Š” MVM1 : ๋น ๋ฅธ ์›€์ง์ž„ ์‹ ํ˜ธ)

b. ๋‹ค๋ฅธ ํ•˜๋‚˜๋Š” 20์ดˆ ๋™์•ˆ์˜ ์›€์ง์ž„์–‘์„ ๋‚˜ํƒ€๋‚ด๋Š” MVM20

- ์ด ์‹ ํ˜ธ๋“ค์€ ์ตœ๋Œ€๊ฐ’์œผ๋กœ ์ •๊ทœํ™”๋˜์—ˆ์œผ๋ฉฐ, motion signal์˜ avg, area, variance, and entropies๊ฐ€ ๊ณ„์‚ฐ๋จ.

- ๋น ๋ฅธ ์›€์ง์ž„ ๋น„์œจ, ์ฆ‰ MVM1์ด 20์ดˆ window ์‹ ํ˜ธ์˜ ์ž„๊ณ„๊ฐ’์ธ 10%๋ฅผ ์ดˆ๊ณผํ•˜๋Š” 10 ๋ถ„ ์ด๋‚ด์˜ ์ƒ๋Œ€์  ์‹œ๊ฐ„๋„ ๊ณ„์‚ฐํ•จ. 

 

 

+ Classify

์˜์‚ฌ ๊ฒฐ์ • ํŠธ๋ฆฌ ๊ธฐ๋ฐ˜์˜ ์ ์‘ํ˜• ๋ถ€์ŠคํŒ…(AdaBoost) ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์‚ฌ์šฉ. 

๋น„๊ต๋ฅผ ์œ„ํ•ด k-nearest neighbors(KNN) ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ support vector machine (SVM) ์‚ฌ์šฉ

 

 

*baseband ์‹ ํ˜ธ, ์ด์ง„์ˆ˜ ๊ฐ’์ด 0 ๋˜๋Š” 1์ธ ๋””์ง€ํ„ธ ์‹ ํ˜ธ

๊ธฐ์ €๋Œ€์—ญ์ด๋ž€ ํ‘œํ˜„์„ ์“ฐ๋ฉฐ, ์›์ฒœ์‹ ํ˜ธ ์˜์—ญ์„ ๋งํ•จ.

๋ฐ˜์‘ํ˜•