#235 AudioDSP

Build and test an arduino DSP circuit based on the pedalSHIELD by electrosmash.

The Build

Notes

I’ve wanted to try some digital signal processing on the Arduino/AVR for a while. When I found the pedalSHIELD by electrosmash, it seemed like an excellent way to bootstrap some experimentation.

This project was initially a breadboard-build of a circuit based on the pedalSHIELD, tested with a simple boost/volume control circuit based on the pedalSHIELD examples. I wanted to see if it was even half-way decent. When that worked out, I put it on protoboard and did a bit more work on the software.

I also purchased the genuine article from electrosmash as a way of supporting the creators and also getting a standard unit for comparison - see LEAP#247 pedalShieldUno for more on that.

Input Processing

I first built the circuit according to the pedalSHIELD schematics, but found that the input chain was prone to clipping and distorting a guitar input signal. I’d prefer a clean signal in, so that any effects are due to the DSP algorithms and later stages, so I made a few modifications:

adjusted to ~50kΩ input impedance
biased the signal to a DC offset of ~ VCC/2
removed the 6.8nF capacitor/filter
changed R4 from 4.7kΩ to 10kΩ to avoid clipping at minimum attenuation (RV1)

Input/Output Gain Stage

The most suitable OpAmp I had on hand was an LMC6482 CMOS Dual Rail-To-Rail Input and Output Operational Amplifier. This prove to work just fine, and with very low power consumption too - all the external circuitry to the Arduino is only drawing about 0.5mA in full operation.

DAC Output Processing

Since the Arduino lacks a precision digital-to-analog output, the output audio waveforms are generated using a neat trick with PWM and RC filter.

It’s possible to use one of more PWM signals, but I’m using two as in the pedalSHIELD design. Basically the two PWM ouputs are set to the high and low bytes of the desired frequency, then mixed with an RC ladder/filter. This reconstitutes (approximately) the desired analog waveform.

Details and more links are available in the article: Configure Arduino UNO PWM outputs to play audio.

For two PWM signals, the resistor selection is R10=4.7kΩ, and R11 = 1.2MΩ (2^n * R10)

With C7=10nF, corner frequency is around 3.38kHz. This might explain why I’m seeing performance radically drop off over 4kHz (see next section on Performance).

Performance

Bottom line:

it works! It’s pretty amazing to jam away in the knowledge that the signal from the guitar is passing through an Arduino before getting to the amp
performance starts to drops off significantly from about 4kHz and by 8kHz it is down in the noise.

I’d like to think it is possible to get better higher frequency performance from the circuit - ideally up to 20kHz - but that needs a bit more investigation. The processor should be capable, but I think the limitation is in the PWM signal generation and mixing/filtering in the output stage.

In the following scope traces:

CH1 - the input signal (FX in)
CH2 - the final output signal (FX out)

Here’s a trace of a 1kHz 200mV signal at minimum input attenuation and default (mid-point) volume boost:

sine_1k

And in the FFT of the output there’s a strong fundamental:

sine_1k_fft

At 2kHz the signal is attenuated:

sine_2k

And the 2kHz output FFT shows more noise and harmonics

sine_2k_fft

At 4kHz the signal is significantly attenuated:

sine_4k

And the 4kHz output FFT shows more noise and harmonics

sine_4k_fft

By 8kHz the signal is pretty much destroyed:

sine_8k

And the 8kHz output FFT shows the signal pretty much drowned by noise and harmonics

sine_8k_fft

Attempting to improve the Frequency Response

With C7=10nF, corner frequency is around 3.38kHz. This might explain why I’m seeing performance radically drop off over 4kHz (see next section on Performance).

At 4kHz, the signal looks like this:

sine_4khz_103

With C7=1nF, the corner frequency is around 33.86kHz, so I was hoping for a better response.

But at 4kHz, the signal shows somewhat less attenuation, but at the expense of an increase in noise:

sine_4khz_102

So no magic bullet. I replaced C7=10nF for the remainder of my tests (I actually put some pin headers on the board so I can swap C7 at will).

Code

I’ve refactored the core routines into a private AudioDspDriver library, so it can be shared amongst some examples. I may split this out as a stand-alone library at some point.

See the following for the examples I’ve tested so far:

Boost is a simple clean boost/volume control
Distortion is a simple clipping distortion effect
Crunch is a distortion effect using asymmetrical clipping based on Schetzen formula
SuperCrunch is a distortion effect using symmetrical clipping based on Schetzen formula