Convolution works by scaling an impulse response according to the value of each audio sample. The processing power required to perform this convolution algorithm is much higher when longer impulse responses are used (adding one sample to the impulse response means that an extra multiplication and addition is needed to compute the output). For this reason, Liquid Mix truncates its impulse response to 300 samples. At 48kHz this is 6.25ms which is fine for high frequencies. However, at lower frequencies, the impulse response is longer, and the Q of the filter is higher. Therefore, the ring in the impulse response is longer and again impulse response needs longer to be to characterise it.

An impulse response of 6.25ms is long enough to convey a period of a sine wave at 160Hz but impulses tend to be made of a combination of decaying sine waves. What happens is one of two things; if the EQ has a lot of 160Hz content in the impulse response, we are truncating the impulse when it is close to zero. In this situation, truncating the impulse response has little effect on the accuracy on the emulation.

In other cases where the impulse response of the EQ has content below 160Hz, the impulse is truncated when the impulse response in not close to zero. These sudden changes in the impulse response during convolution cause all frequencies below 160Hz to be comb filtered. Liquid Mix attempts to compensate for this. However this compensation is not perfect and leads to filters that may not have the correct frequency response at low frequencies or that are asymmetrical.

## Comments