Preamp

[Monolith]

That is useful info . From a system point of view, this can define system signal to noise ratio as long as you do the testing in an electrical quiet setting. If your scope or measuring device includes an FFT function you can look at the signal in the frequency domain which makes determining the frequency of the noise components trivial. Unfortunately, using the FFT on most scopes is tough for an inexperienced user. If anyone is interested in trying I can give some general tips. Most are not obvious...

I would very much like to get your tips on FFT. My pre-amp signal input must have a minimum bandwidth of 1MHz.
Some years ago, I used SPECTRUMLAB, to look at my signal. Attached is a screen shot. You can see the FFT. It shows distinct noise at the 60Hz mains and 120Hz harmonic.
Unfortunately, as it uses the sound card of the computer, it's frequency range is limited.
I am going to search for this software again, hope it works with Windows 10.

[Monolith]

My thoughts on signal to noise, hope some are close to correct. First, need to compare peak to peak not RMS. Sweep coil over target and miss the next sweep. Have a pulse .1 to.2 seconds long. RMS is going to approach zero. Peak to peak white noise typically 6 times RMS. When does the target signal disappear. Searched minimum signal to noise, 3 to 1 was a suggested value. Example: scales, lsd 1 gram, noise 0 to 5 grams. Add 5 grams, displays 5 to 10 grams. If the weight was added and removed when the no weight reading was 0 the reading would be in the noise band. Signal needs to be greater than noise to have signal reading greater than noise reading every time. 3mv noise doesn't tell me what I need to know. Eric was probably referencing a test point on a particular detector which would if I could calculate the gain. Noise and signal should be referenced to the input(coil volts). I'm thinking if I could detect a 1 uvolt change in coil volts that would be good. Did another schematic for a 1 and 10 uvolt test signal. Reply #67 didn't work, missed the obvious reason why. Appreciate comment if I missed something this time.

I admire your systematic, methodical approach. To be able to detect a 1uVolt target response would be excellent.

The 3mV noise that Eric was talking about was taken at the end of signal processing, before the signal voltage is converted to the audio output. The same applies to the screen shots I posted. So this is the accumulation of all noise. The input and pre-amp noise can then only be inferred, by changing the front-end, coil, shielding, op-amp etc.

[6666]

http://www.qsl.net/dl4yhf/spectra1.html

try here

[Monolith]

http://www.qsl.net/dl4yhf/spectra1.html

try here

Thanks, I got it and it seems to work fine on Windows 10

[green]

That is useful info . From a system point of view, this can define system signal to noise ratio as long as you do the testing in an electrical quiet setting. If your scope or measuring device includes an FFT function you can look at the signal in the frequency domain which makes determining the frequency of the noise components trivial. Unfortunately, using the FFT on most scopes is tough for an inexperienced user. If anyone is interested in trying I can give some general tips. Most are not obvious...

Always wanting to learn something. We used FFT where I worked, sometimes it made sense to use FFT and sometimes not. Could you give an example using FFT to determine signal to noise for a PI detector?

[Old cart]

FFT usage and hints

FFT's are a mathematical way to convert from time domain (amplitude vs time) to frequency domain (amplitude vs frequency).While these are just different views of the same signal (object) each domain has advantages when it comes to looking for different things. If the signal is complex waveform it is difficult to decompose it into the various frequency components. In the case of our PI's the signal is ideally a rectangular pulse of width t. The Spectrum of this ideal pulse would be a Flat line starting at DC with the first zero in the response at 1/t. There would be successively smaller cosine shaped humps after that and successive zeros in the response at 2,3,4,5...times 1/t.

There used to be some great online calculators to help you visualize what any time domain signal looks like in the frequency domain. I suggest you do a google search for fourier component calculator to find one that works it's your computer. This wil help give Aquino understanding about how FFT's work. A general search for FFT's yields lots of information but much of it is very math heavy.

However there are few basic rules for using the FFT's in any device, like a scope that has this function.

1. Time and frequency are inverse domains so good time resolution wil give poor frequency resolution.
2. Set the scope record length to around 10000 points.
3. Set the sample rate so that it is around 4 times the highest frequency component you expect to observe. For example if you want to look at all the noise coming out of the first stage of the preamp and want to see all the signal to say 250Khz set the sample rate to 1 Ms/S.
4. Make the signal fill the screen vertically to the extent this is possible.
5. Turn on FFT. You should see a signal in the frequency domain. Many scopes will allow you to place cursors on the spectrum and measure amplitudes at various frequencies. Use haning or hamming windowing if this is adjustable on your scope and if the signal contains periodic components. No windowing is necessary if the signal contains broadband noise.
6. You may have to adjust the sweep speed slightly to show the best frequency domain detail.
7. Since FFT's are just mathematical transforms of the time series of the data they have limitations.
a. You wil not be able to se signal frequencies higher than 1 / sample rate. ( 1 MS shows signals up to 500KHz BUT it is best to ignore anything great than 250 KHz in this example)
b. The best resolution you can get is the sample rate /2 / .5 times the record length so with the setting above this becomes
1MS/s / 2= 500KHz/ 5000 points = 100 Hz. So with thes setting it would not be possible to differentiate the difference between 50 Hz and 100Hz, they would just blob together on the display.
c. The best vertical dynamic range you can get is approximately 6 x the number of bits your a to d converter has ( 8 or 48 dab) in a general purpose scope.

Using a sound card with a PC can give much broader range in amplitude but not in the frequency range since most sound cards are limited to 96KHz sample rate and give a frequency range of 48KHz. This is usually adequate to look at the output after the integrator since it severely band limits the signal. I use a pice of great software from Croatia cal ARTA. Even the free version if fully functional except you can not save waveforms or spectrum displays.

[Old cart]

Thanks, I got it and it seems to work fine on Windows 10

that is nice software. It will also let you try out various filters and see the effect on the noise.

[Old cart]

Always wanting to learn something. We used FFT where I worked, sometimes it made sense to use FFT and sometimes not. Could you give an example using FFT to determine signal to noise for a PI detector?

See if the comments above are adequate. If not just ask...

[green]

See if the comments above are adequate. If not just ask...

Thanks for reply #81. Shows me how to get a good FFT. What I'm missing in the PI example is how to use the FFT plots. It still looks easier to see if the peak to peak signal is grater than the peak to peak noise. The FFT could show high 60 Hz noise when it could have a low 60 Hz peak to peak because it's continuous.

[Old cart]

Thanks for reply #81. Shows me how to get a good FFT. What I'm missing in the PI example is how to use the FFT plots. It still looks easier to see if the peak to peak signal is grater than the peak to peak noise. The FFT could show high 60 Hz noise when it could have a low 60 Hz peak to peak because it's continuous.

Got it. Since I have no plots it is hard to give you a precise example. However I think you make a good point. For a simple quick test you can simply measure the ratio of the P-P noise and compare that to the P-P signal. That is not likely to be very accurate as the ear is not sensitive to peak but rather more likely to respond to the sound power or voltage squared power. More importantly the FFT on most scopes is a power spectrum so more closely responds to what you would hear. Lastly the FFT will give you reasonably accurate frequencies which can be very important when you are trying to figure out where the noise is coming from. For example if there is a peak at 10KHz it is likely to be coming from the charge pump. If there is a rising noise spectrum below a couple of hundred Hz it is likely caused by an op amp near the input that has excessive 1/f noise. Knowing this makes it much easer to isolate the problem.

One process could go like this.

1. Disable the transmitter and measure the preamp output noise with the coil replaced by a resistor of equivalent value.
2. Exchange op amps or adjust the restor values around the op amp to maintain constant gain but reduce noise. Generally to lower Johnson noise you lower the value of the resistors. Unfortunately if you are using protection diodes you can not lower the input resistor much below 1000 ohms. FFT's are very useful at this point. If noise at specific frequencies is present try to figure out where it is coming from, for example via the supply lines or perhaps due to board contamination or maybe even RFI being rectified in the input diodes.
3. Turn the transmitter back on and connect the coil. Measure the noise using the same set up as above. By using the scope trigger it should be possible to only capture data when the transmitter is off and the signal has decayed. In this way you are only measuring noise, but at more of a system level. If your scope has the ability to measure the RMS value of the spectrum you can measure the effective noise power broadband.
4. Repeat the noise reduction efforts but this time you wil be seeing the effects of recovery of the big current pulse drawn by the coil, power supply instability and other added noise.
5. Move the trigger forward so that the coil flyback is captured to measure MOSFET Avalanche noise on the preamp output.
6. Record and think about all noise effects and try to get them as low as possible.
7. Move to the output of the integrator and repeat, once again trying to get the lowest noise. You may find that the noise level varies as you move the sample points and widths. There is lots of room for tuning here.
8. Keep moving forward in the signal path one stage at a time. Since the bandwidth is going down in each stage you may have to slow down the scope sweep speed to capture the frequency range you are interested in. Since you are reducing the noise at each step in the process the noise should be going down BUT each stage typically has gain so noise is likely to go up. Absolute value is NOT the important thing, what you are trying to do is to lower the value from where you started. It is also important to note that you always get to see the time domain waveform at every step which a help to ensure none of you noise lowering efforts causes other problems.
9. Somewhere after the integrator you can introduce a target into the equation. By noting the target response you may find you can reduce the stage bandwidth. Knowing that noise goes down with the sq root of the bandwidth helps. Note also that with a target you should adjust threshold so that you have normal audio response. This lets you detremine if some of the noise is coming back from the audio stage, which are often pretty crude and can introduce noise.

NOTE, some more advanced scopes allow you to average the frequency domain, on the FFT. This is very useful to determine if the noise is synchronous withe the transmitter ( noise wil not go down with averaging) or asynchronous ( noise will go down ).

One last point it is good dea to practice using the FFT's on a KNOWN signal, like the calibrator signal. This 1 KHz square wave, which has all odd harmonics ( 1,3,5,7,9,11KHz) out to maybe 11 KHz, should be sampled at around 50 to 100 KS/s. You wil be able to learn about how to best use the FFT on your scope as well as what effect adjusting the controls have.

[Sean_Goddard]

Have a look at this; http://www.cirrus.com/en/pubs/proDat...2-34-BS_F3.pdf

With two you would have a sample rate of 96KHz X 4 = 384KHz and 23Bit resolution. What if each of the samples was taken in the following manner Pulse 1 = samples 1,2,3,4, pulse 2 = samples 5,6,7,8 etc like is done with a fast scope?

Any milegage in this or do we need faster? If so what sample rate do we need. Can we sample the decay curve at 4 points and use them as a reference by using each ADC channel per point?

In a project I was saw a long time ago, radio signals were being decoded from all over the world. The signal average level was -123dB and these were certainly readable in terms of clarity of audio after processing. Considering "Earth Noise" is at around -127dB I think that is pretty good going. Just shows you what is possible with the correct hardware, unfortunately I can't post any schematics here as the project methods used were classified.

I can only sit back and watch now as you guys have this in hand and are teaching me a few things.

Does anyone else think it would be a good idea to go DUAL COIL? At least we could eliminate a lot of crud from the Tx from getting into the Rx signal (hopefully).

[Monolith]

that is nice software. It will also let you try out various filters and see the effect on the noise.

|Thanks Old cart, for the tips and explanation for FFT. Very useful.

[Old cart]

Have a look at this; http://www.cirrus.com/en/pubs/proDat...2-34-BS_F3.pdf

With two you would have a sample rate of 96KHz X 4 = 384KHz and 23Bit resolution. What if each of the samples was taken in the following manner Pulse 1 = samples 1,2,3,4, pulse 2 = samples 5,6,7,8 etc like is done with a fast scope?

Any milegage in this or do we need faster? If so what sample rate do we need. Can we sample the decay curve at 4 points and use them as a reference by using each ADC channel per point?

In a project I was saw a long time ago, radio signals were being decoded from all over the world. The signal average level was -123dB and these were certainly readable in terms of clarity of audio after processing. Considering "Earth Noise" is at around -127dB I think that is pretty good going. Just shows you what is possible with the correct hardware, unfortunately I can't post any schematics here as the project methods used were classified.

I can only sit back and watch now as you guys have this in hand and are teaching me a few things.

Does anyone else think it would be a good idea to go DUAL COIL? At least we could eliminate a lot of crud from the Tx from getting into the Rx signal (hopefully).

This technique is called interleaving. It can certainly be done but is quite tricky, I would think, at the hobbiest level. The samples have to be placed very accurately, within a few pS of the right place to avoid errors when the waveform is spliced together. Also some sort of calibration is necessary between the individual A to D converters so they all match perfectly. Otherwise the number of effective bits ( the measure of how much waveform distortion is introduced) will be decline. This is particularly important at higher frequencies. I think direct conversion,using the A to D to replace the integrators and beyond, is a great idea on a high performance software defined detector. It has not been used in commercial designs because the high performance converters are quite expensive. Also software cost are high. I like the idea of using audio converters as there has been a lot of development work done on them. The particular device you mention have serval neat features like programmable gain and high impedance differential inputs. It also has integrated digital filters. I like separate Tx and RX coils.

[Teleno]

So what about the preamp?

[Old cart]

So what about the preamp?

Good point, but everything here is at least about the preamp measurement, selection and optimization!