CW ESR denoising when triplets meet wavelets Boris Dzikovski, - - PowerPoint PPT Presentation

CW ESR denoising – when triplets meet wavelets

Boris Dzikovski, ACERT

Denoising with wavelets

• Popular in signal processing since 1990s
• Wavelet transform leads to a sparse

representation for many real-world signals.

• Wavelet coefficients which are small in

value are considered noise and can be removed without affecting the signal.

• Wavelet denoising results in better

approximations of the original signal compared to e.g. Fourier filtering etc.

• The new method with intelligent

thresholding allows for recovering signals even for seemingly hopeless cases.

Nicer and smoother, yes. But is it more informative?

3340 3350 3360 3370 3380 3390

Magnetic Field, G

• 0.02

0.02

Magnitude

Can we extract more information by denoising? And the answer is:

Yes!

Extracting polarity/mobility parameters from Noisy aiso = 17.18G, τ = 1.2×10-11s Denoised aiso = 17.25G, τ = 1.7×10-11s “No noise” spectrum aiso = 17.25G, τ = 1.8×10-11s

Tempo in water

3280 3300 3320 3340 3360 3380

Magnetic Field, G noisy long accamulation denoised D D1

Extracting D1/D parameter from Noisy 0.507 Denoised 0.414 “No noise” spectrum 0.413

Tempo in water/glyc, 100K

N O

3290 3300 3310 3320 3330 3340 3350 3360

Magnetic Field, G noisy denoised long accumulation Tempol in CCl4, deoxygenated

Extracting super hyperfine structures

N O OH

3140 3160 3180 3200 3220

Magnetic Field, G

• 0.2
• 0.1

0.1 0.2

Magnitude

3140 3160 3180 3200 3220

Magnetic Field, G

• 0.2
• 0.1

0.1 0.2

Magnitude

Miracles are not guaranteed

Tempone in water – the ability to extract single very fine features meets its limits.

SNR ~ 1 SNR ≤ 0.25 N O O

But miracles happen – sometimes.

3300 3320 3340

Magnetic Field, G noisy denoised long accumulation

di-p-tert-butyl-phenyl nitroxide in toluene, degassed

N O

And such spectra are not uncommon:

N N

• Pyrazine anion

N N CH3 H3C + +

2Cl-

.

Methyl Viologen

H2C C C CH2

[ ] - Butadien ion

Denoising is able to reveal underresolved hyperfine splitting features:

Same experimetal sample, Two different amplitude modulation: M=0.2G M=1G Detail components 2-3-4 kept but approximation component put to zero

3300 3320 3340

Magnetic Field, G noisy denoised reference signal

Spin trapping, a lot of weak signals. Much better with denoising.

+ •OH

• In some spin-trapping experiments without denoising the

very presence of a trapping adduct is uncertain. Denoising allows for reliable identification and quantification of the reaction products.

• 18
• 16
• 14
• 12
• 10
• 8
• 6
• 4
• 2

• 1
• 0.5

ENDOR – another case of traditionally weak signals

Both pulse (Mims and in particular Davies) and CW ENDOR require very long accumulation time due to low sensitivity.

A lot of opportunities to apply denoising techniques at various stages of spectral processing: From denoising single transients to the CW-style denoising of the resulting ENDOR spectrum

Biological tyrosyl radical – Mims Endor PD-Tempone in D2O/d8-glyc with some H2O content, Davies ENDOR

CW ESR = multiple scans (usually)

Choosing “good” scans using wavelet denoising – “scan sorting”.

PC spin label in liposomes of mixed lipids

Should we average denoised scans?

Magnetic Field, Tesla 95 GHz Magnetic Field, Tesla 95 GHz Every scan denoised Denoised after accumulation

Better to denoise after averaging…

Real or Ghost?

Testing spectral features by using alternating set of scans with following denoising.

Still here Odd scans Even scans

Thank you! Welcome to ACERT!