Audio sw - Removing DC offset from sampled data

Projects / Audio Switching  Original post date:06/28/2016

Normally, one would use a DC blocking cap in the analog world. I wanted to do that, but I need the low pass filter even more in the ADC front end.

I am sampling a bipolar signal with an ADC by offsetting the input, so at the end of the day, I am still dealing with DC offsets.

Calibration

The DC offset of the system is measured and subtracted off. So far, that's what I have done and it work great so far. I have notices the source only have a few mV of offsets. The front end opamp circuits would probably add another few mV. Normally this is not a big problem, but the offsets are affecting how small a signal I can detect in my code. There are a few ways of removing DC offsets in a sample.

Averaging

The samples are averaged. The average is the DC offset. That works for post processing when you have accessing to a lot of samples. It gets a bit tricky when I have small blocks of data. I have tried doing this, but this seems to mess up my volume.

High Pass filter

A high pass filter can be used as DC block by having the filter frequency set close to 0Hz. This is what I found on a datasheet sample circuits. I tweaked the values until I get something that works well with parts I have. i.e. no overshooting, fast settling, right DC offset.

LTSpice simulation of a high pass filter

This is the filter characteristic.

Frequency/Phase response of filter

This is what it does in the Time Domain. The blue trace is the input waveform with a DC offset of -1V and the green one is the output.

Output vs input waveform

I can get away with messing around tweaking the values in analog circuits by feel without knowing the math. Unfortunately, math is behind all that DSP stuff in making a digital filter. So far I have seen lots of Z transforms or Matlab only and they looks all Greek to me.

I found some online calculator and examples for the filter coefficient, but the numbers are very close to unity, so that it would requires some fixed point calculations. I am not sure if the truncation errors would be significant when I am dealing with small numbers without a lot of resolution. I tried one example code on a forum and the offset for my small signal comes out to be 0.

Trying to do this with fancy DSP math might not be what I want.

Cheats

The daughter card has 4 analog signals. Other than part counts/board space, there are no reasons why the audio detection can't be done in analog circuits and simply send back 2 digital signals. This was the original design before I try to do everything in the uC.

Reality Check

May be I am over thinking. I am going to look into this.

Averaging is what I am using in the new version of firmware. Found my problem in the volume calculation. I should have removed the DC offset before calculating the volume. This actually works well without fuzz.

I haven't reverse engineered the $0.30 Class D amplifier yet, but back of my mind thinks that the gain in the Class D amplifier might be set too high. For a low power amplifier, the chip has up to 24dB voltage gain (~15x) depending on the external components.  I have since reduced the gain of the speaker amplifier to about 3.5X by replacing the external components here.