Designing Analog-Modeled Saturation and Wave-Shaping Algorithms for VST3 Plugins

Pure digital audio can sound sterile and harsh because it lacks the gentle harmonic coloration and natural peak compression of vintage hardware. Creating authentic analog-style saturation requires non-linear transfer curves coupled with rigorous anti-aliasing filter stages.

1. The Mathematics of Hyperbolic Tangent Soft-Clipping

The hyperbolic tangent function, tanh(x), is the quintessential mathematical model for analog triode tube distortion. For small inputs, tanh(x) ≈ x, producing completely linear output. As the signal approaches unity amplitude, the transfer curve smoothly saturates, compressing peaks and introducing warm odd harmonics:

inline float processTanhSaturation(float inputSample, float driveGain) {
    float driven = inputSample * driveGain;
    // Approximating tanh for CPU efficiency:
    // Pade approximant or direct fast-math intrinsic
    return std::tanh(driven);
}

2. The Aliasing Menace: Why Oversampling is Mandatory

Non-linear wave-shaping generates harmonic components that extend far beyond the original signal bandwidth. According to the Nyquist-Shannon sampling theorem, any generated harmonic frequency exceeding half the sample rate ($f_s / 2$) reflects backwards into the audible spectrum as dissonant inharmonic distortion, known as aliasing.

To eliminate aliasing, the input signal must be upsampled (typically 4x or 8x) using a steep linear-phase polyphase FIR filter before saturation, and subsequently filtered and decimated back to the original host sample rate.

// JUCE DSP Oversampling Block
juce::dsp::Oversampling<float> oversamplingEngine { 2, 2, juce::dsp::Oversampling<float>::filterHalfBandPolyphaseIIR };

void prepare(const juce::dsp::ProcessSpec& spec) {
    oversamplingEngine.initProcessing(spec.maximumBlockSize);
}

void process(juce::AudioBuffer<float>& buffer) {
    juce::dsp::AudioBlock<float> block(buffer);
    juce::dsp::AudioBlock<float> oversampledBlock = oversamplingEngine.processSamplesUp(block);

    // Apply non-linear wave-shaping at 4x sample rate
    for (size_t ch = 0; ch < oversampledBlock.getNumChannels(); ++ch) {
        auto* data = oversampledBlock.getChannelPointer(ch);
        for (size_t i = 0; i < oversampledBlock.getNumSamples(); ++i) {
            data[i] = std::tanh(data[i] * 2.5f);
        }
    }

    oversamplingEngine.processSamplesDown(block);
}

3. Summary

Analog warmth is not magic; it is the deliberate application of non-linear mathematics combined with high-order oversampling to preserve high-frequency clarity.