EMI Noise Reduction Design Ideas for PI Detectors

[Carl-NC]

Good idea to look at it with the RX coil shorted and not shorted. Shows that internal noise is not insignificant. I assume "differentiator out" is the output of the SAT just past the demod opamp. If so, then all of this is low frequency noise because the demod opamp has a very low frequency cutoff, typically 10-20Hz. By throttling the preamp you are reducing the HF noise that gets aliased to LF noise in the demod. In the final screens you are down to just LF noise. If you have EFE cancellation running then that will cancel a lot of LF noise, it is generally a comb filter. You might try turning it off.

I would measure the noise right off the preamp before demodulation. Then you get a look at raw noise before the demod moves it all around.

[moodz]

Some scope pictures. Recordings were made on the kitchen table. Don't see any LF(<100Hz)noise if I did it right. Anyone see anything that doesn't make sense?

capacitor was used to change amplifier cutoff frequency

Hey green nice shots .. can you show us the circuit ?

[green]

Hey green nice shots .. can you show us the circuit ?

A schematic of the integrator circuit I posted awhile back, some of the component values might have changed. Scope pictures, swinging a coin across end of a fig8 Rx coil. The detector schematic is in pieces, will attach a full schematic after I put it together if you are interested.

Going to do a couple more tests that Carl suggested first.

Don't remember if GB was on causing the nickel and quarter to be opposite polarity swinging across one end or if they were swung across opposite ends.

[green]

Good idea to look at it with the RX coil shorted and not shorted. Shows that internal noise is not insignificant. I assume "differentiator out" is the output of the SAT just past the demod opamp. If so, then all of this is low frequency noise because the demod opamp has a very low frequency cutoff, typically 10-20Hz. By throttling the preamp you are reducing the HF noise that gets aliased to LF noise in the demod. In the final screens you are down to just LF noise. If you have EFE cancellation running then that will cancel a lot of LF noise, it is generally a comb filter. You might try turning it off.

I would measure the noise right off the preamp before demodulation. Then you get a look at raw noise before the demod moves it all around.

Maybe what you suggested. "differentiator out" B attachment reply #33. Shows that internal noise is not insignificant. Amplifier is noisy or fig8 Rx is cancelling most of the noise seen at differentiator out. What should I expect to see at amplifier out? Amplifier out noise is higher when coil not shorted, maybe at frequencies not aliased into integrator response. Recording of differentiator out with integrator I in connected to analog ground not bad but would go off scale when I waved my hand over circuit causing air flow, maybe amplifier drift.

[green]

Hey green nice shots .. can you show us the circuit ?

Normal PI except for differential amplifier and figure8 RX coil. Any suggestions on how to make it better?

[green]

Wondered if increasing amplifier cutoff frequency would effect noise. Repeated 1MHz and tried 3MHz. Amplifier noise increased some, not much difference at differentiator out. Don't know if using 3MHz would be better, no difference or worse?

First and third picture in the row has Rx shorted, second and forth picture Rx connected to Rx coil. 1000pps

[drninja]

Might be the base current noise of the BJTs. It gets differentiated by the coil inductance, and its gets worse with frequency. Try replacing the RX coil with compact inductors with different inductances and check if noise changes with inductance value.

[green]

I had a 1 turn trim pot to adjust sample rate(pps)to minimize differentiator out noise. Replaced it with a 20 turn trim pot and monitored differentiator out on the scope as I adjusted the pot. Adjustment range, 898Hz to 1015Hz. Frequency display on scope not real stable so Hz numbers are ball park. Slowly adjusted pot and recorded frequency where peak noise increased at least 4 times. Might have missed some points since a small pot rotation gets output back to normal. 900, 915, 928, 942, 957, 974, 990 and 1008Hz display caused noise to increase at least 4 times. Probably aliasing on some frequency, anyway to tell what frequency?

Tx coil disconnected. Rx_200mm figure8, 5us delay, 10us sample.

[drninja]

I have no idea how else but to hook a spectrum analyzer on the frontend amplifier output. Maybe try to modulate the sample rate with something to mitigate it (spread spectrum)?

[green]

In another thread it was suggested to adjust sample rate half way between multiples of line frequency to minimize aliasing noise. 990Hz as an example (16*60=960, 17*60=1020). Line frequency isn't exactly 60 Hz all the time, 60.2Hz*16=963.2 sampled 960 times/second would give a 3.2Hz signal. I get aliasing close to 900, 930, 960 and 990Hz. Any thoughts why? My EF sample isn't half way between target samples(about 300us, 700us)could that be the reason I see aliasing half way between multiples of line frequency?

My scope has FFT but so far scope doesn't seem to have the resolution needed.

[drninja]

Well yeah, 930Hz is not divisible by 60 but its second harmonic is. Sampling this way, mathematically is multiplication of the signal and sampling waveforms in the time domain, which becomes convolution of their spectra in the complex frequency domain. Your sampling waveform has its fundamental at the pulse repetition rate, and its shape is encoded in its harmonic content. If you adjust the EF to be centered, you get rid of the even harmonics in the spectra of your sampling waveform (certain symmetries in time domain are directly linked to the presence and/or ratio of even/odd harmonics in the frequency domain) which ends up in less things convolving with each other.

Edit: I think you should filter the 60Hz out somehow before the sampler and/or try spread spectrum, centering the EF permanently might not be good idea from a performance point of view.

[Altra]

In another thread it was suggested to adjust sample rate half way between multiples of line frequency to minimize aliasing noise. 990Hz as an example (16*60=960, 17*60=1020). Line frequency isn't exactly 60 Hz all the time, 60.2Hz*16=963.2 sampled 960 times/second would give a 3.2Hz signal. I get aliasing close to 900, 930, 960 and 990Hz. Any thoughts why? My EF sample isn't half way between target samples(about 300us, 700us)could that be the reason I see aliasing half way between multiples of line frequency?

My scope has FFT but so far scope doesn't seem to have the resolution needed.

I some times sync my scope to "Line" instead of TX. This allows you to see the mains drift and lock on frequencies.

[green]

Well yeah, 930Hz is not divisible by 60 but its second harmonic is. Sampling this way, mathematically is multiplication of the signal and sampling waveforms in the time domain, which becomes convolution of their spectra in the complex frequency domain. Your sampling waveform has its fundamental at the pulse repetition rate, and its shape is encoded in its harmonic content. If you adjust the EF to be centered, you get rid of the even harmonics in the spectra of your sampling waveform (certain symmetries in time domain are directly linked to the presence and/or ratio of even/odd harmonics in the frequency domain) which ends up in less things convolving with each other.

Edit: I think you should filter the 60Hz out somehow before the sampler and/or try spread spectrum, centering the EF permanently might not be good idea from a performance point of view.

Thanks
Wasn't thinking. Was sampling target sample 990 samples/second but sampling noise 1980samples/second. 1980/60=33. Need to see what centering EF sample does.

[green]

Thanks
Wasn't thinking. Was sampling target sample 990 samples/second but sampling noise 1980samples/second. 1980/60=33. Need to see what centering EF sample does.

Need to see what centering EF sample does Tried with my bipolar circuit(same time between samples). Unipolar circuit has target and EF sample. At Tx 1000pps, 2000 samples/second EF trails target sample by about 700us.
Bipolar circuit: no aliasing at multiples of line frequency, aliases half way between. Tx frequency should be a multiple of line frequency.
Unipolar circuit: aliases at multiples of line frequency and half way between. Tx frequency should be a multiple of line frequency+-1/4 line frequency.

Listed frequency(samples/second)that differentiator out(SAT out?)was at least 4times greater than than normal(no aliasing of line frequency).

[Vadim18]

Anybody tried to randomise TX rate in small range, say -+5% pps?
Say TX is 1000 with variable -+50 pps.
I think it will help to escape from interference, or at least interference will not be constant since pps is random.