Windowed-sinc filters are used to separate one band of frequencies from another. They are very stable, produce few surprises, and can be pushed to incredible performance levels.
These exceptional frequency cutoff characteristics are obtained at the expense of poor performance in: a) the time domain (meaning they will always introduce latency), b) excessive ripple and c) overshoot in the step response.
In the image below you can see how the signal pulse (Oscillator 280Hz) is delayed as the number of “Taps” at a filter frequency of 5 kHz increases, and the amount of “ringing” or ripple on the pulse waveform also increases.
Increasing TAPs increases the latency:
TAPs and latency.
The SINC Lowpass filter is a linear-phase FIR Filter. The number of “taps” specifies the number of coefficients, more TAPs means increased filter cut-off steepness as seen below however, increasing the number of TAPs introduces more latency. In SynthEdit however there is latency compensation. The module is watching for any change to the default value of the ‘Taps’ pin. The module then uses this value to calculate how much latency compensation it requires and passes that value to the host via the ‘SetLatency’ method. Latency is measured in sample frames.
The module reports this latency to SynthEdit to enable PDC (Plugin Delay Compensation). PDC hides the effect of latency through the clever use of delay lines. You can literally think of all FIR/SINC filters as a multitap delay with no feedback, all the taps are spaced 1 sample apart, then gain is applied per tap and then all taps are added together. So for a 171 tap delay, you have 170 delays, 171 gains (like level adjust modules) and 170 adds. It’s already well optimised with SSE2, as it’s doing 4 calculations at a time.
Resonance/Q/Feedback.
SINC filters do not include a feedback path, so have no feedback, resonance, or “Q” control plug. The are intended to be used as a filter with a very steep low/high pass cutoff characteristic, rather than for colouration of sound.
Increasing the number of TAPs and the effect on frequency roll-off.
Note about 0 Hz filter cut-off:
Although this filter will allow you to set a cut-off frequency of 0 Hz, you cannot use a filter frequency of 0Hz, this is is an ‘illegal‘ value. You will get quite loud clicks and pops, along with “glitching”.
Most SynthEdit filters are “clipped” internally so that end-users don’t input wrong values (this comes at a slight CPU cost of course), usually limited to just above 0Hz and just below the Nyquist frequency.
Why 0Hz is illegal is easiest to explain with a simple 6dB/Octave lowpass. If we set the lowpass to 100Hz, then 200Hz (2nd octave) will be filtered by -6dB, 300Hz (3rd Octave) by -12dB, 400Hz(4th Octave) by -18dB and so on, hence the “6dB/Octave” name. Now try and do the same thing with 0Hz….what is an octave above 0? 0 multiplied by 0 is still 0. It’s the same as dividing by zero.
To prevent loud pops, clicks and glitching you must limit the lowest cut-off frequency to 14 Hz.
Just be sure to limit your patch memory values to prevent illegal values.
Other than this the SINC filters are suitable for fast modulation of the cut-off frequency. Note that there is no resonance/feedback on these filters.
Changing TAP value.
One point to note with SINC filters is that when you change the TAP value the audio engine has to reset for the recalculation process.
TAP Value and filter cut-off slope.
Although we can vary the cutoff slope on SINC filters by altering the TAP number, this bears no relation to the more familiar dB/Octave slope, so unfortunately you cannot say that a certain TAP number is equivalent to a particular dB/Octave cutoff slope. This is due to the DSP structure, and the way these filters work, they are a using a different method of filtering which bears little or no mathematical resemblance to the more traditional analogue filter emulations.
Leave a Reply