This came up recently in

another thread
.

Here are some pointers:

**Too short an impulse may impact the low frequency accuracy:**
If the impulse is very short (i.e. less than 5000 samples at 44.1 KHz samplerate) then you might be losing some accuracy in the low frequencies. Whether or not this matters is subjective from impulse to impulse. Most (probably all) convolver implementations use a mathematical 'FFT' or Fast Fourier Transform. This technique takes a time domain impulse and transforms it into the frequency domain where it is applied to the signal (also transformed into the frequency domain). If the FFT length is short, then the frequency 'buckets' become more course at lower frequencies, hence you may see inaccuracy at lower frequencies in the resulting frequency response. Note that the higher the sample rate, the more pronounced the effect. Through experimentation, I've found that at 44.1KHz sample rate, an impulse of 5000 samples or greater is usually enough to minimise any inaccuracies.

The S-GEAR Scuffham impulses are between 6000 and 9000 samples in length at 44.1 sample rate. At higher sample rates the impulses are longer. In fact the impulses are are probably a bit longer than they need to be.

*How much room response is captured in the impulse?*
In the earlier thread I mentioned that the appropriate length of an impulse can depend on how much of the room has been captured in the impulse, and how much or that effect you want in the resulting signal.

The most common measurement technique is to run an exponential sine sweep through the system under test and deconvolve the result with the inverse filter. This method does not attempt to separate the room response from the equipment response. You will capture both time domain effects (reflections and echoes) and frequency domain effects (frequency response and frequency-dependent reverberation). Truncating the impulse will remove some of the time domain effects of the room.

If the impulse was captured with

**no room effects**, then you might summise that the maximum required IR length might be about 2200 samples (at 44.1) to capture frequencies down to 20Hz. However, given the short impulses issue described above, you may hear a significant difference between the 2200 and the 5000 sample impulses. In reality, most impulses have are not totally free of room effects, so they may have a somewhat longer tail anyway.

The Scuffham impulses in S-GEAR were recorded in an acoustically treated professional studio - one with a pleasantly damped sound. I didn't truncate the tails.

Also, note that the ProConvolver MKII allows you to import any length IR, but still truncates impulses internally to a maximum of 20,000 samples. More than enough for any cabinet response.

I hope the above info helps.

Mike