Dynamics processing: Compressor/Limiter, part 1

Lately I’ve been busy developing an audio-focused game in Unity, whose built-in audio engine is notorious for being extremely basic and lacking in features. (As of this writing, Unity 5 has not yet been released, in which its entire built-in audio engine is being overhauled). For this project I have created all the DSP effects myself as script components, whose behavior is driven by Unity’s coroutines. In order to have slightly more control over the final mix of these elements, it became clear that I needed a compressor/limiter. This particular post is written with Unity/C# in mind, but the theory and code is easy enough to adapt to other uses. In this first part we’ll be looking at writing the envelope detector, which is needed by the compressor to do its job.

An envelope detector (also called a follower) extracts the amplitude envelope from an audio signal based on three parameters: an attack time, release time, and detection mode. The attack/release times are fairly straightforward, simply defining how quickly the detection responds to rising and falling amplitudes. There are typically two modes of calculating the envelope of a signal: by its peak value or its root mean square value. A signal’s peak value is just the instantaneous sample value while the root mean square is measured over a series of samples, and gives a more accurate account of the signal’s power. The root mean square is calculated as:

rms = sqrt ( (1/n) * (x₁² + x₂² + … + x_n²) ),

where n is the number of data values. In other words, we sum together the squares of all the sample values in the buffer, find the average by dividing by n, and then taking the square root. In audio processing, however, we normally bound the sample size (n) to some fixed number (called windowing). This effectively means that we calculate the RMS value over the past n samples.

(As an aside, multiplying by 1/n effectively assigns equal weights to all the terms, making it a rectangular window. Other window equations can be used instead which would favor terms in the middle of the window. This results in even greater accuracy of the RMS value since brand new samples (or old ones at the end of the window) have less influence over the signal’s power.)

Now that we’ve seen the two modes of detecting a signal’s envelope, we can move on to look at the role of the attack/release times. These values are used in calculating coefficients for a first-order recursive filter (also called a leaky integrator) that processes the values we get from the audio buffer (through one of the two detection methods). Simply stated, we get the sample values from the audio signal and pass them through a low-pass filter to smooth out the envelope.

We calculate the coefficients using the time-constant equation:

g = e ^ ( -1 / (time * sample rate) ),

where time is in seconds, and sample rate in Hz. Once we have our gain coefficients for attack/release, we put them into our leaky integrator equation:

out = in + g * (out – in),

where in is the input sample we detected from the incoming audio, g is either the attack or release gain, and out is the envelope sample value. Here it is in code:

public void GetEnvelope (float[] audioData, out float[] envelope)
{
    envelope = new float[audioData.Length];

    m_Detector.Buffer = audioData;

    for (int i = 0; i < audioData.Length; ++i) {
        float envIn = m_Detector[i];

        if (m_EnvelopeSample < envIn) {
            m_EnvelopeSample = envIn + m_AttackGain * (m_EnvelopeSample - envIn);
        } else {
            m_EnvelopeSample = envIn + m_ReleaseGain * (m_EnvelopeSample - envIn);
        }

        envelope[i] = m_EnvelopeSample;
    }
}

(Source: code is based on “Envelope detector” from http://www.musicdsp.org/archive.php?classid=2#97, with detection modes added by me.)

The envelope sample is calculated based on whether the current audio sample is rising or falling, with the envIn sample resulting from one of the two detection modes. This is implemented similarly to what is known as a functor in C++. I prefer this method to having another branching structure inside the loop because among other things, it’s more extensible and results in cleaner code (as well as being modular). It could be implemented using delegates/function pointers, but the advantage of a functor is that it retains its own state, which is useful for the RMS calculation as we will see. Here is how the interface and classes are declared for the detection modes:

public interface IEnvelopeDetection
{
    float[] Buffer { set; get; }
    float this [int index] { get; }

    void Reset ();
}

We then have two classes that implement this interface, one for each mode:

A signal’s peak value is the instantaneous sample value while the root mean square is measured over a series of samples, and gives a more accurate account of the signal’s power.

public class DetectPeak : IEnvelopeDetection
{
    private float[] m_Buffer;

    /// <summary>
    /// Sets the buffer to extract envelope data from. The original buffer data is held by reference (not copied).
    /// </summary>
    public float[] Buffer
    {
        set { m_Buffer = value; }
        get { return m_Buffer; }
    }

    /// <summary>
    /// Returns the envelope data at the specified position in the buffer.
    /// </summary>
    public float this [int index]
    {
        get { return Mathf.Abs(m_Buffer[index]); }
    }

    public DetectPeak () {}
    public void Reset () {}
}

This particular class involves a rather trivial operation of just returning the absolute value of a signal’s sample. The RMS detection class is more involved.

/// <summary>
/// Calculates and returns the root mean square value of the buffer. A circular buffer is used to simplify the calculation, which avoids
/// the need to sum up all the terms in the window each time.
/// </summary>
public float this [int index]
{
    get {
        float sampleSquared = m_Buffer[index] * m_Buffer[index];
        float total = 0f;
        float rmsValue;

        if (m_Iter < m_RmsWindow.Length-1) {
            total = m_LastTotal + sampleSquared;
            rmsValue = Mathf.Sqrt((1f / (index+1)) * total);
        } else {
            total = m_LastTotal + sampleSquared - m_RmsWindow.Read();
            rmsValue = Mathf.Sqrt((1f / m_RmsWindow.Length) * total);
        }

        m_RmsWindow.Write(sampleSquared);
        m_LastTotal = total;
        m_Iter++;

        return rmsValue;
    }
}

public DetectRms ()
{
    m_Iter = 0;
    m_LastTotal = 0f;
    // Set a window length to an arbitrary 128 for now.
    m_RmsWindow = new RingBuffer<float>(128);
}

public void Reset ()
{
    m_Iter = 0;
    m_LastTotal = 0f;
    m_RmsWindow.Clear(0f);
}

The RMS calculation in this class is an optimization of the general equation I stated earlier. Instead of continually summing together all the  values in the window for each new sample, a ring buffer is used to save each new term. Since there is only ever 1 new term to include in the calculation, it can be simplified by storing all the squared sample values in the ring buffer and using it to subtract from our previous total. We are just left with a multiply and square root, instead of having to redundantly add together 128 terms (or however big n is). An iterator variable ensures that the state of the detector remains consistent across successive audio blocks.

In the envelope detector class, the detection mode is selected by assigning the corresponding class to the ivar:

public class EnvelopeDetector
{
    protected float m_AttackTime;
    protected float m_ReleaseTime;
    protected float m_AttackGain;
    protected float m_ReleaseGain;
    protected float m_SampleRate;
    protected float m_EnvelopeSample;

    protected DetectionMode m_DetectMode;
    protected IEnvelopeDetection m_Detector;

    // Continued...

public DetectionMode DetectMode
{
    get { return m_DetectMode; }
    set {
        switch(m_DetectMode) {
            case DetectionMode.Peak:
                m_Detector = new DetectPeak();
                break;

            case DetectionMode.Rms:
                m_Detector = new DetectRms();
                break;
        }
    }
}

Now that we’ve looked at extracting the envelope from an audio signal, we will look at using it to create a compressor/limiter component to be used in Unity. That will be upcoming in part 2.

Audio Resampling: Part 1

Resampling in digital audio has two main uses.  One is to convert audio into a sampling rate needed by a particular system or engine (e.g. converting 48kHz audio to the required 44.1kHz required by CDs).  The second is to avoid aliasing during signal processing by raising the Nyquist limit.  I will be discussing the latter.

Lately I’ve been very busy working on improving and enhancing the sound of ring modulation for a fairly basic plug-in being developed by AlgoRhythm Audio (coming soon).  I say basic becuase as far as ring modulation goes, there are few DSP effects that are simpler in theory and in execution.  Simply take some input signal, multiply it by a carrier signal (usually some kind of oscillator like a sine wave), and we have ring modulation.  The problem that arises, however, and how this connects in with resampling, is that this creates new frequencies in the resulting output that were not present in either signal prior to processing.  These new frequencies created could very well violate the Nyquist limit of the current sampling rate during processing, and that leads us to resampling as a way to clamp down on aliasing frequencies that can be introduced as a result.

Aliasing is an interesting phenomenon that occurs in digital audio, and in every sense of the word is an undesired noise that we need to make sure does not pollute our audio.  There are many resources around that go into more detail on aliasing, but I will give a brief overview of it with some audio and visual samples.

Aliasing occurs when there are frequencies present in a signal that are greater than the Nyquist limit (half of the sampling rate).  What happens in such a case is that the sampling rate is not high enough to properly capture (sample) the high frequency of the signal, and so the frequency “folds over” and creates aliases that are mirror images of the original frequency.  Here, for example, is a square wave at 4000Hz created using 20 harmonics at a sampling rate of 44.1kHz (keep in mind that 4000Hz is the fundamental frequency, and that square waves contain many additional frequencies above that depending on how many harmonics were used to create it, so in this case Nyquist is still being violated):

4000Hz square wave with aliasing

Notice the low tone below the actual 4000Hz frequency.  Here is the resulting waveform of this square wave that shows us visually that we’re not sampling fast enough to accurately reproduce the waveform.  Notice the inconsistencies in the waveform.

Now, going to the extreme a bit, here is the same square wave sampled at a rate of 192kHz.

4000Hz square wave with no aliasing

It’s a pure 4000Hz square wave tone.  Examining the waveform of this square wave shows us that the sampling rate was more that adequate to reproduce this signal digitally:

Not all signal processing effects are susceptible to aliasing, and certainly not to the same degree.  Because ring modulation produces additional inharmonic frequencies, it is a prime example of a process that is easily affected by aliasing.  Others include distortion and various other kinds of modulation techniques (especially when taken to the extreme).  However, ring modulation with a sine wave is generally safe as long as the frequency of the sine wave is kept  low enough because sine waves have no harmonics to them, only the fundamental.  Introducing other wavetypes into this process, however, can quite easily bring about aliasing.

Here is an example of ring modulation with a sawtooth wave sweeping up from about 200Hz to 5000Hz.  As it glides up, you will be able to hear the aliasing kicking in at around 0:16 or so.  The example with no aliasing has been upsampled by 3X before processing, then downsampled by 3X back to its original rate.

Ring modulation with a sawtooth sweep, with aliasing

Ring modulation with sawtooth sweep, with no aliasing

So how does upsampling and downsampling work?  In theory, and even to some extent in practice, it’s very straightforward.  The issue, as we will see, is in making it efficient and fast.  In DSP we’re always concerning ourselves with speedy execution times to avoid latency or audio dropouts or running out of memory, etc.

To upsample, we insert 0-valued samples in between every L-th (our upsampling factor) original sample.  (i.e. upsampling by 3X, [1, 2, 3, 4] becomes [1,0,0,2,0,0,3,0,0…])  However, this introduces aliasing into our audio so we need to interpolate these values so that they “join” the sample values of the original waveform.  This is accomplished by using an interpolating low-pass filter.  This entire process is known as interpolation.

Dowsampling is much the same.  We remove every M-th (our downsampling factor) sample from the original signal.  (i.e. downsampling by 2X, [1,2,3,4,5] becomes [1,3,5…])  This process does not introduce aliasing, but we do need to make sure the Nyquist limit is adhered to at the new sampling rate by low-pass filtering with a cutoff frequency at the new Nyquist rate.  This entire process is known as decimation.

Fairly straightforward.  Proceeding with this method, however, would be known as brute force — generally not a good way to go.  The reason why becomes clearer when we consider what kind of low-pass filter we need for this operation.  The ideal filter would be one that would brick-wall attenuate all frequencies higher than Nyquist and leave everything else untouched (thus preserving all the frequencies and tonal content of our original audio).  This is, alas, impossible, as it would require an infinitely long filter kernel.  The function that would implement this ideal filter is the rectangular function.

The rectangular function.

By taking the Fourier transform of the rectangular function we end up with the sinc function, which is given by:

y(x) = sin (x) / x, which becomes y(x) = sin (πx) / πx for signal processing.

Graph of the sinc function, sin (πx) / πx

The sinc functions trails on for infinity in both directions, which can be seen in the graph above, so we need to enforce bounds around it by applying a window function.  Windowing is a method of designing FIR filters by essentially “surrounding” a function (in our case the sinc) by the window, which enforces bounds so that we can properly derive a filter kernel for use in calculations.  The rectangular function shown above is a type of window, but as I mentioned, infinite slope is a deal-breaker in audio.

The Blackman window, given by the function

w(i) = 0.42 – 0.5 cos(2πi / M) + 0.08 cos(4πi / M),

where M is the length of the filter kernel, is a good choice for resampling because it offers a good stop-band attenuation of -74dB with good rolloff.  Putting this together with the sinc function, we can derive the filter kernel with the following formula:

Windowed-sinc kernel formula*

where f_c is the normalized cutoff frequency.  When i = M/2, to avoid a divide by zero, h(i) = 2f_c.  K is a constant value used to achieve unity gain at zero frequency and can be ignored while calculating the kernel coefficients.  After all coefficients have been calculated, K can be found by summing together all the coefficients, h(i), and then dividing each by the resulting sum.

* Source: Smith, Steven W., “The Scientist and Engineer’s Guide to Digital Signal Processing”, Chapter 16.

Now that we have the filter design, let’s consider the properties of the FIR filter and compare them to IIR filter designs.  IIR filters give us better performance and attenuation at lower orders, meaning that they execute faster and perform better with fewer calculations than FIR filters.  However, IIR filters are still not powerful enough, even at slightly higher orders, to give us the performance we need for resampling, and trying to push IIR filters into very high orders can make them unstable and/or susceptible to quantization error due to the nature of recursion.  IIR filters also do not offer linear phase response.  FIR filters are the better choice for these reasons, but the unfortunate drawback is that they execute slowly due to being implemented by convolution.  In addition, they need to be pushed to high orders to give us the performance needed for attenuating aliasing frequencies.

However, the order of the interpolating low-pass filter can be negotiated based on the frequency content of the audio signal(s) involved.  If the audio is sufficiently oversampled, it will not contain enough frequencies near Nyquist, and as such a lower order filter can be used with a gentler rolloff without adversely affecting the audio and attenuating actual frequencies in the signal.  There are plenty of cases where we just don’t know the frequency content of the audio signals involved in processing, however, so a strong filter may be needed in these cases.  Here is a graph of a 264-order window-sinc filter (in other words, a filter kernel of length 265 including the sample x(0)):

264-order window-sinc low-pass filter frequency response (cutoff frequency at 10kHz, resulting in a transition band of 882Hz)

With this in consideration, it can be easy to see that convolving a signal with a 264-order FIR filter is computationally costly for real-time processing.  There are a number of ways to improve upon this when it comes to resampling.  One is using the FFT to apply the filter.  Another interesting solution is to combine the upsampling/downsampling process into the filter itself, which can further be optimized by turning it into a polyphase filter.

The theory and implementation of a polyphase filter is a fairly long and involved topic on its own so that will be forthcoming in part 2,where we look at how to implement resampling efficiently.