A question about ADCs
![pik33](https://forums.parallax.com/uploads/userpics/026/nFBJPVUJK5HCS.jpg)
in Propeller 2
I did a simple test with ADC and DAC. What I hear is a white noise at about -60 dB. DACs are silent: put 0 there and no noise can be hear so it is ADC which generate the noise.
This means I got only about 10 effective bits.
I used pins 0 and 1 ad inputs using cinch socketd on an accessory board.
Is it normal? Or I am doing something wrong here?
Also the input impedance is very high
Can it be set lower in the program or... add the external resistor at input?
' ADC and DAC CON _clkfreq = 300000000 adcpinl = 0 adcpinr = 1 dacpinl = 6 dacpinr = 7 pub start() coginit(7,@mixer,0) cogstop(cogid()) DAT org mixer wrpin adc_config, #adcpinl wxpin #%10_1001, #adcpinl dirh #adcpinl wrpin adc_config, #adcpinr wxpin #%10_1001, #adcpinr dirh #adcpinr wrpin dac_config,#dacpinl wxpin a512,#dacpinl wrpin dac_config,#dacpinr wxpin a512,#dacpinr dirh #dacpinl addpins 1 setse1 #%001<<6 + adcpinl loop rdpin x,#adcpinl rdpin y,#adcpinr sub x,diffx1 add diffx1,x sub x,diffx2 add diffx2,x sub y,diffy1 add diffy1,y sub y,diffy2 add diffy2,y shr x,#11 shr y,#11 wypin x,#dacpinl wypin y,#dacpinr waitse1 'wait for new period jmp #loop 'loop adc_config long %0000_0000_000_100011_0000000_00_11000_0 dac_config long %0000_0000_000_10111_00000000_01_00011_0 diffx1 long 0 diffx2 long 0 diffy1 long 0 diffy2 long 0 affff long $FFFF x long 0 y long 0 p1 long 0 p2 long 0 a512 long 512
Comments
Now there is less noise. I have to understand how these ADCs work....
The simplest mode is 2^n-clock SINC2 sampling, set by 'WXPIN #%00_nnnn'.
You don't even need to use those filteted ADC smart pin modes. If you do use them, your sample only becomes stable on the third sample from start.
In some cases, I just use the count-highs smart pin mode. It gives you a good sample on the first measurement, but without any filtering.
The 1x input mode has an impedance of ~540k to VIO/2. The 100x input mode has an impedance of ~5.4k over a 40 mV range.
When using the "filtering" modes, be sure to use one of the workarounds for register size mismatch - https://forums.parallax.com/discussion/comment/1510074/#Comment_1510074
EDIT: I now see you're ANDing after the SAR instructions. I'm not sure SAR is suitable there, although it may not matter because all extended bits are stripped by the AND anyway.
I applied this 5 higher bit mask, nothing changed.
The noise depends on pin and CPU clock frequency. This is not linear: there are silent frequencies and noisier frequencies. The character of noise also changes with cpu frequency, and also non linear so I can't tell "higher (or lower) frequency noises less/more.
Pin 0 is more silent than pin 1 in this environment. When the frequency is high (>300) and gain >1 a constant high frequency can be heared. At 200 MHz there is a strange, granular noise on pin 0
To be continued![:) :)](https://forums.parallax.com/resources/emoji/smile.png)
Right, it makes no difference to your low noise issue but that mask you're now using is important to remove severe numerical glitches that only happen under some odd-ball case. It can definitely happen out of the blue without one of the preventative fixes.
Yes, the 2^n sinc is the simplest: rdpin and nothing else. About the high impedance: I had a jack in one hand and a P2 board with adc->dac program running with only output connected in another hand: the music played out of the headphone jack![:) :)](https://forums.parallax.com/resources/emoji/smile.png)
How to combine "count highs" with adc?
That's the plain single accumulator as per prop1 days. And can be called sinc1. It's present in the collection of counter modes of the smartpins:
- For internal clocked: %01111 AND !Y[0] = Count A-input highs
- For external clocked: %01100 = Count A-input positive edges when B-input is high
For audio applications, I got the best results (less noise) from the simplest solution. Noise dac noises less than pwm dac in this configuration. The ADC is supposed to be 14 bit in this configuration and it seems it is something like 14 bit: I didn't measure the noise level yet.
I am thinking now about a noise shaper applied directly on adc raw bit stream to move the noise out of accoustic band before integrating anything. I don't know if it is possible or achievable on a P2.
I wonder whether the high clock rates are affecting the ADC performance? Perhaps via self heating too?
This thread (and paper) from RJSM might be worth a look
https://forums.parallax.com/discussion/169602/characterizing-p2-eval-analog-performance
This ADc2DAC works for several hours now, no anomalies detected. P2 is cold, but it has near nothing to do, one cog waiting for ADC pin ready.
The topic was interesting, this can partially explain why several clock frequencies gives less noise (PLL jitter) and why the noise DAC can noise less (the same LDO for ADC and DAC)
In our early experimentation with the ADCs, we found that 1/f noise was the one problem that we could not eliminate. Maybe you would have some ideas about how to improve things. Remember, there's no problem in using 2 or 4 ADCs, or even 8, at once, if you want higher resolution. Lots of strange things are possible.
Chip, can you please explain what we have to do about 1/f noise? I know it can't be avoided completely but can it at least be minimized? I've read somwhere that 1/f noise (as the name already tells) would become infinite at f=0Hz (DC) and long term offset drift can also be explained as 1/f noise.
So if we make the sampling time longer or put a low pass filter over multiple samples, is there a limit where further filtering does not result in lower noise? Or is there an optimum where shorter sampling time would cause more quantisation noise and longer sampling time more 1/f noise?
It seems that shorter measurements are better, because there's less time involved.
For a noise free audio input we don't need 16bit absolute precision as the noise will be masked by louder parts of the music. What we need is to get rid of the noise in silent parts. These ADCs have a controlled gain, so maybe controlled switching between gains can do this. Or use 3 adcs, 1,10 and 100 x and combine their output together. Calibration problems may occur.
About 70 dB SNR is what a good vinyl record can output. At this level maybe even a simple DNL solution can do the trick.
So many ideas and still not enough knowledge.