10 releases

0.2.12-alpha.0 Apr 7, 2023
0.2.11-alpha.0 Dec 19, 2022
0.2.5-alpha.0 Jun 21, 2022
0.2.4-alpha.0 Mar 14, 2022
0.1.42-alpha.0 Oct 27, 2021

#19 in #non-linear


Used in surge-synthesizer

GPL-3.0 license

430KB
9K SLoC

surgefilter-nlstates

NonlinearStatesFilter is a Rust module implementing a nonlinear feedback filter, which can be used in various audio signal processing applications. The module provides functionality for calculating filter coefficients using the Audio EQ Cookbook method, as well as different types of saturators for the filter stages. The filter stages themselves are implemented using the Direct Form I topology, which is a common structure for digital filter implementations.

The filter can be tuned to different cutoff frequencies using the clamped_frequency function, which ensures that the frequency remains within a certain range to prevent unstable behavior.

Overall, NonlinearStatesFilter provides a flexible and customizable way to implement nonlinear feedback filters in Rust, with support for different types of filters and saturators.

<coeff_make>

The coeff_make function in NonlinearStatesFilter calculates the filter coefficients using the Audio EQ Cookbook method. This involves determining the resonance (reso) and frequency (freq) of the filter, then calculating the different coefficients using the following formulas:

q = (reso ^ 3) * 18.0 + 0.1

w0 = 2 * π * freq / sr

sin(w0) = wsin

cos(w0) = wcos

alpha = wsin / (2 * q)

a0 = 1 / (1 + alpha)

a1 = -2 * wcos * a0

a2 = (1 - alpha) * a0

b0 = (1 - cos(w0)) * a0 / 2

b1 = (1 - cos(w0)) * a0

b2 = (1 - cos(w0)) * a0 / 2

The resulting coefficients are used in the Direct Form I topology to implement the filter.

<clamped_frequency>

The clamped_frequency function in NonlinearStatesFilter ensures that the cutoff frequency remains within a certain range to prevent unstable behavior. It takes a pitch value and uses a TunerHandle to convert it to a frequency value. The frequency is then clamped to a range between 5 Hz and 30% of the sample rate, which is returned as the output. The function uses the following formula to calculate the frequency from the pitch:

freq = n2p(pitch + 69) * MIDI_0_FREQ

<process_quad>

The process_quad function in NonlinearStatesFilter implements the Direct Form I topology to process a single sample of input data through the nonlinear feedback filter. The function takes a QuadFilterUnitState and an input value, and returns the output of the filter for that input value. The function applies different types of saturators and calculates the filter coefficients for each stage, using the coefficients calculated by the coeff_make function. The Direct Form I topology is implemented using a set of z-1 state registers and coefficient registers, which are updated for each stage of the filter.

The test_process_quad_allpass function is an alias of process_quad, with the difference that it is specialized for the Allpass filter type. This allows for a more efficient implementation of the Allpass filter, which has a simpler coefficient calculation than the other filter types.

The test_process_quad_allpass_complex function is an alias of process_quad, with the difference that it is specialized for the Complex Allpass filter type. This allows for a more efficient implementation of the Complex Allpass filter, which uses complex coefficients.

The test_process_quad_bandpass function is an alias of process_quad, with the difference that it is specialized for the Band

52> what is the Direct Form I topology?

The Direct Form I is a common topology used to implement digital filters. It is a type of transposed direct form filter structure, which means that it is derived from the canonical direct form filter structure.

The Direct Form I topology represents a recursive filter using a series of additions and multiplications. It consists of a set of delay elements, or "z^-1" blocks, which represent the delay of the input signal, and a set of "b" and "a" coefficient blocks that implement the filter transfer function.

The transfer function of a digital filter in Direct Form I can be represented mathematically as:

Y(z) = b0*X(z) + b1*X(z^-1) + b2*X(z^-2) - a1*Y(z^-1) - a2*Y(z^-2)

where X(z) and Y(z) represent the input and output signals, respectively, and b0, b1, b2, a1, and a2 are the filter coefficients.

The Direct Form I topology is a relatively simple and efficient way to implement recursive filters, but it can be sensitive to coefficient quantization errors, and it is not well-suited for implementing high-order filters due to its numerical stability issues.

53> are there other related topologies? are there drawbacks to this one?

Yes, there are several other related topologies for implementing digital filters. Two of the most commonly used ones are Direct Form II and Transposed Direct Form II.

Direct Form II is similar to Direct Form I, but the order of the operations is different. In Direct Form II, the input is first fed into a set of feedback delays, and then the feedback coefficients are applied to these delayed values. The output of the feedback section is then fed into a set of feedforward delays, and the feedforward coefficients are applied to these delayed values to produce the final output. Direct Form II is often preferred over Direct Form I because it can be more numerically stable for high-order filters.

Transposed Direct Form II is similar to Direct Form II, but the order of the operations is reversed. In Transposed Direct Form II, the feedforward and feedback sections are swapped, so that the input is first fed into the feedforward section and then into the feedback section. Like Direct Form II, Transposed Direct Form II can be more numerically stable than Direct Form I for high-order filters.

One potential drawback of Direct Form I is that it may not be as numerically stable as other topologies for high-order filters. This is because the feedback coefficients are applied to the input signal before it is delayed, which can lead to large intermediate values that may cause numerical issues. Additionally, Direct Form I requires twice as many delays as Direct Form II or Transposed Direct Form II, which can increase memory requirements. However, for lower-order filters, Direct Form I can be a simple and efficient way to implement a digital filter.

54> could you write a bullet list of pros and cons for each of these topologies you listed? feel free to add more rows for more topologies if there are some other interesting ones

Sure! Here's a bullet list comparing some of the pros and cons of the Direct Form I, II, and Transposed Direct Form II topologies:

Direct Form I:

  • Pros:
    • Simple and easy to implement
    • Low computational cost
  • Cons:
    • Numerical noise amplification

    • Requires more memory compared to other structures

    • Can exhibit overflow and limit cycle issues

Direct Form II:

  • Pros:

    • Lower numerical noise amplification than Direct Form I

    • Can be implemented with fewer memory elements than Direct Form I

  • Cons:

    • Requires more computational resources than Direct Form I

    • Can be less stable than Direct Form I

Direct Form II Transposed:

  • Pros:

    • Can have a lower noise floor compared to Direct Form I and Direct Form II

    • Uses the least amount of memory of the three structures

  • Cons:

    • Can be more difficult to implement than Direct Form I and Direct Form II

    • Requires more computation than Direct Form I

Cascade Form:

  • Pros:
    • Stable and robust
    • Low numerical noise amplification
  • Cons:
    • Higher computational cost than Direct Form I and Direct Form II

    • Can be more difficult to implement and optimize than other structures

Parallel Form:

  • Pros:

    • Low computational cost
    • Can be implemented with multiple cascaded biquad stages for increased accuracy
  • Cons:

    • Numerical noise amplification
    • More complex to implement than Direct Form I and Direct Form II

56> thanks -- what else should we keep in mind while using this crate? how does it differ from surgefilter-nlfeedback?

Here are some things to keep in mind while using surgefilter-nlstates:

  • This crate provides an implementation of a non-linear state variable filter (NSLVF) using the Direct Form I topology.

  • The NSLVF is a type of filter that can create non-linear distortion effects by adding feedback loops to a basic state-variable filter structure.

  • The filter coefficients are calculated using the same algorithm as surgefilter-nlfeedback, but the two crates use different filter topologies.

  • This crate is optimized for performance on modern processors and makes use of SIMD instructions to process multiple samples in parallel.

  • The filter can be used in a variety of different applications, including audio processing, image processing, and more.

Compared to surgefilter-nlfeedback, which provides a similar filter implementation but uses a different topology, surgefilter-nlstates has some advantages and disadvantages:

Advantages:

  • The Direct Form I topology used in surgefilter-nlstates is generally more efficient than the Direct Form II topology used in surgefilter-nlfeedback.

  • surgefilter-nlstates is optimized for performance on modern processors and makes use of SIMD instructions to process multiple samples in parallel.

Disadvantages:

  • The Direct Form I topology used in surgefilter-nlstates can be more prone to numerical instability than the Direct Form II topology used in surgefilter-nlfeedback.

  • surgefilter-nlstates does not provide the same range of filter types as surgefilter-nlfeedback. For example, surgefilter-nlfeedback includes a comb filter and a state-variable filter, which are not available in surgefilter-nlstates.

Dependencies

~11–20MB
~301K SLoC