Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

Version 1 Next »

The Audio Weaver Filters folder lists over 60 filters. They have been broken down according to user needs, with the folder labels Adaptive, Calculated Coeffs, Controllable, High Precision, Raw Coeffs, and list the most commonly used filters. The Adaptive folder contains the LMS module, an adaptive filter with tracking capabilities. For those users less experienced with designing filters, the Calculated Coeffs filters take in frequency information, Q, Gain, and type, similar to tuning a filter in a DAW. Users with more DSP background can use the Raw Coeffs filters to tune filters with mathematical information. The most frequently used filters are the ButterworthFilter (highpass, lowpass, allpass), SecondOrderFilterSmoothed, with 20 different filter types, and the SecondOrderFilterSmoothedCascade: multiple 2nd order filters in series.

Adaptive (LMS)

The LMS filter predicts the FIR of a system whose transfer function is not given. It’s input and output adapt or “predict” what the system response is. Filter weights are updated over time based on mu speed, higher numbers being the faster update speed. Higher numtaps give higher chance to converge with the optimum filter weight(meaning less error). The error can be tracked realtime with the errorSignal output. The module comes with an option to output the predicted “coeffs”. The following system shows white noise being ran through a 10 point FIR. The LMS will predict the FIR coefficients, and sinks will display the error and coeff function.

This sink shows the coeff prediction.

The LMS is trying to predict this FIR response.

The Error2 display shows a value of -125 dB, which means that our signal is very accurate. The sink to the right displays this as well.

Filters with Calculated Coeffs

Audio Weaver has a several filters with built-in design equations. These filters are implemented using Biquad or BiquadCascade modules behind the scenes and the filter coefficients are computed by the design equations based on high-level filter specifications. The design equations use the sample rate on the input wire when computing coefficients. The following sections describe each of the calculated coeffs modules.

Allpass Pair

The Allpass Pair module creates a pair of allpass filters with the special property that their sum and difference form a doubly complementary highpass-lowpass filter pair. When used in conjunction with the Sum and Difference module (available in the Math folder) this Allpass Pair module can be used to construct more complex structures such as N-way crossovers and filter banks. The following design mixes one channel of white noise into two “bands” of white noise using this technique. The sink is an FFT showing the frequency response of the signal.

Audio Weighting Filters

The AudioWeighting module is located under Filters/Calculated Coeffs This module applies different standard audio weighting to a signal. The available weightings are selectable from the inspector and include: A, B, C, and D-weighting, as well as ITU468, LeqM, and ITU 1770. The WeightingFilter is used for noise measurements and broadcast loudness applications. The following frequency responses display each weighting’s filter.

Crossover Filter

A crossover is a special type of filter that splits a signal into multiple bands, while sustaining a total gain of 0 dB. It will boost the level of one frequency band as the other drops to compensate to 0 dB gain. This behavior is shown in the figure below.

Crossovers are used in loudspeaker applications to separate signals into different frequency bands to be output via woofers, mid-range, and tweeter speakers. They are implemented using ButterworthFilter (odd-order) or Linkwtiz-Riley (even-order) filters. Crossover filters can be made manually using individual filters. By cascading filters and applying an allpass filter during other lane filter stages (shown below), more crossover points can be added and the signal split into more frequency bands, while retaining the unity gain property.

Alternatively, use the crossover module which contains all of the needed filters. The Crossover Filter module allows the user to set the type of filter, number of output bands, and filter order, specified in the module’s properties. Crossovers are used for separating different frequency bands of a signal. The example below demonstrates a crossover module being used to split a signal into a high band above 250 Hz and a low band below 250 Hz. The sum of the levels of the two bands is always 0 dB. As the input frequency changes near 250 Hz, one channel’s level drops and the other smoothly increases to compensate to 0 dB.

This is the same behavior as using two ButterworthFilter filters:

allows the configuration of 2 or more output channels. The module will then be drawn with 3 output pins. The top pin is the low frequency; the center pin is the mid-range, and the bottom pin is the high frequencies. The inspector shows two cutoff frequencies: between the low and mid-range; and between the mid-range and high frequencies. For example, if implementing a 3-way loudspeaker crossover, configure it as:

Emphasis Filter

The EmphasisFilter module implements a pre-emphasis or de-emphasis, used for noise reduction. The cutoff frequency is specified by the time constant tau, which is set in the inspector. The examples below show emphasis and de-emphasis filters with 75 microsecond time contants.

Graphic EQs

The GraphicEQ module splits up a signal into different bands and independently attenuates or amplifies each band. Under module names and arguments, the number of bands and the order of each filter are set. The bands are logarithmically spaced across the Nyquist frequency of the input signal. Each band’s gain can be set in the inspector. This EQ can automatically adjust its bands based on the lowEdge and highEdge arguments. After setting this to the desirable range, change the resetCenterFreqs flag to ‘1’. This will recalculate the bands, throwing away all slider data (so change the bands before tuning).

  • No labels