Scribd
Upload
94 views

The Use of Digital Signal Processing Techniques in Audio Effects

This document summarizes key aspects of digital signal processing techniques for audio effects as described by Robert Chidlaw, Chief Scientist at Source Audio LLC. It discusses hardware requirements, advantages of digital over analog, control techniques using multiplication instead of variable resistors, numerical precision considerations, and algorithms for filters, transitions between presets, and control feel. It also covers objective measures of performance like noise and distortion, and the need for subjective evaluation by musicians. The overall goal is to describe how to design digital effects circuits that sound great to serious musicians.

Uploaded by

Jadur Rahman
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
94 views

The Use of Digital Signal Processing Techniques in Audio Effects

This document summarizes key aspects of digital signal processing techniques for audio effects as described by Robert Chidlaw, Chief Scientist at Source Audio LLC. It discusses hardware requirements, advantages of digital over analog, control techniques using multiplication instead of variable resistors, numerical precision considerations, and algorithms for filters, transitions between presets, and control feel. It also covers objective measures of performance like noise and distortion, and the need for subjective evaluation by musicians. The overall goal is to describe how to design digital effects circuits that sound great to serious musicians.

Uploaded by

Jadur Rahman
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

The Use of Digital Signal Processing Techniques in

Audio Effects
By Robert Chidlaw, Chief Scientist, Source Audio LLC

Introduction
I was asked to become a founding member of the Source Audio
team, because they knew how important it is to create
guitar effects that actually sound good to serious
musicians. The other team members came from a big chip
company and while they knew electrical engineering, they
did not have the experience of being in the industry.
Through my 20 years of experience as Chief Scientist at
Kurzweil and 40 overall years of tinkering and building
guitar effects and amps, I have learned a few things about
how to make things sound good. I will describe in this
paper some of the key things required to make effects
circuits (especially digital ones) sound great.

Hardware
The first thing we need is a processor with sufficient
computational horsepower to get the job done right. It
also has to be inexpensive enough such that we can sell the
finished product at a reasonable price. Our SA601 chip
meets these goals. It is expressly designed for the
efficient processing of audio signals.

Digital vs analog
I have built these kinds of signal processing circuits
using analog hardware during the 1970’s and 80’s. In many
ways, it’s so much easier in digital. There are no issues
with component tolerances, no stray capacitance between
wires, and no hum. All active components (tubes and
transistors (either singly or integrated into an op amp))
have some intrinsic noise level. There are noise sources
in digital processing, but we can choose to arrange the
details of the computation such that the final resultant
noise is as small as we wish.

Control - variable resistors vs multiplication


In analog, controls are often implemented with
potentiometers, which are variable resistors that provide a
hand adjustable amount of signal attenuation. This
attenuation can be used to control the frequency of an
analog filter circuit. If this control is to be
electronically varied, then a multiplication must be done
between two signals, the control signal and the audio
signal. This is quite a challenge to do in analog without
a lot of cost, noise, or distortion – you can’t minimize
all three simultaneously.

In the DSP world, the multiplication operation is just the


same as what we learned in elementary school. It’s just
done a little faster. Our SA601 chip can do 48 million
multiplications in a second.

Numerical precision
Now we get to the specifics of the DSP arithmetic. Word
size is of critical importance. Every time a
multiplication is done, the word length would have to
double to preserve the exact answer. For a practical
implementation with a fixed word length, we are constantly
discarding data from the least significant portion of each
multiplication result, and thus adding noise. Although the
basic word size in the SA601 is 28 bits, support is
provided for double precision operations such that the
number of extra instructions required is not a problem.
Using such a 56 bit accumulator where needed means that the
unavoidable noise appears at such a low level as to be
negligible.

The SA601 uses fixed point arithmetic, where the binary


point appears five bits down from the most significant bit.
Fixed point arithmetic is cheaper to implement in silicon
than floating point. With floating point, part of the data
word is used to specify a scale factor, which means that
the same amount of precision is always available, for small
numbers or large numbers. Fixed point programming is more
difficult to write, as care must be taken that the
computational values do not get too small (and lose
precision) or too large (and clip.) But in the end it
helps to force an understanding of exactly what is
happening during each computation.

Division and table look-up


If you think back to grade school, you probably remember
that division seemed a lot harder to do than
multiplication. Well, that’s also true for DSP chips.
There is no division instruction in the SA601. If a
division cannot be avoided, then a table of reciprocals is
used. Interpolation is used to get adequate resolution
from a coarse table. At times, a cubic interpolation
polynomial is used to meet the required accuracy.
Table look up is used to evaluate trigonometric, power law,
and exponential functions as well. Direct computation of
these kinds of functions is quite out of the question. It
also permits creation of empirically derived functions,
whose exact shape is tailored to produce good sounding
results.

Algorithms

Filters
Some of the filter algorithms are built using the digital
state variable filter configuration. The various filter
parameters have been adjusted by ear to sound interesting
and be useful. The multi-peak algorithms, with up to three
peaks or dips, have an awful lot of parameters. I can’t
claim to have explored the entire range of possible sounds,
but certainly a good-sized chunk. That was one of the more
interesting and creative parts of the project.

To evaluate these algorithms, there must be testing by


musicians. Algorithms must be judged to sound good and
feel and play right. It is a totally subjective analysis.
During development, we had a lot of musicians come in and
try out various algorithms, and rate them according to how
much they liked them. A number of algorithms got both the
lowest possible and the highest possible scores. It was
very difficult to make the final choices. We tried to come
up with a set that has a good variety of useful sounds.

The “classic” wah, on the other hand, is a digital


implementation of the exact frequency response of the
original wah circuit. It was more an exercise in
mathematics than a question of what sounds right.

Transitions
All transitions from one sound to another are smoothed.
When moving from one preset to another, or adjusting a
front panel control, the resultant changes in the numbers
that govern the computation process ramp smoothly from
their current value to their new value. This avoids clicks
and pops that might otherwise occur.

Control feel
The accelerometer signals must be mapped into control of
the filter parameters in a way that provides a good natural
feel to the musician. This did not turn out to be all that
difficult, but one is faced with interpreting comments by
musicians attempting to describe what they like and
dislike, and mapping those comments into changes in the
code. This is where it is best if the designer is also a
player.

Objective measures
Well, OK, there are some objective measures of performance.
Noise is almost always bad. (Although at my previous job
we put a noise source in a flanger algorithm to simulate
the old noisy analog delay lines. Personally, I don’t use
it.)

Unwanted distortion is bad. In analog, higher order


distortion components generally drop off rapidly. Digital
distortion tends to have only very slowly decreasing
amounts of high order distortion components, unless a lot
of care is taken in how the distortion is simulated.

Aliasing distortion is particularly bad in digital, since


it has no counterpart in analog. The harmonic distortion
components that digital distortion generates march right up
to half the sampling frequency and beyond. The frequencies
beyond half the sampling rate are then reflected back down
again, to lower frequencies that have no audible
relationship to what went in.

There are still open questions as to the audibility and


desirability of the inevitable small amounts of distortion
always present in analog circuits. That’s getting to be a
lot of computation, to add analog-like distortion
everywhere throughout a digital algorithm. At this point,
we get back to subjective evaluations of whether such a
high degree of precision in a model is worth the extra cost
of the computations.

Summary
I don’t feel that digital audio processing needs to have
compromises today. If implemented correctly, it should be
able to meet and exceed all the specs of analog gear. In
addition, there are many things, with many more coming,
that can only be done digitally.

You might also like