AndreasN wrote on Sat, 04 September 2010 14:54 |
Hmmm... That was confusing. The starting portion makes me wonder. It's usually a steady noisefloor followed by an abrupt transition to the impulse at 0ms. Your graph starting climbing upwards at 50ms ahead of the impulse. Do you run room correction or something similar on the signal? Is there maybe something that introduces a lot of distortion? Wrong samplerate? Something something? |
Patrik T wrote on Sat, 04 September 2010 17:27 |
Hi I'll run a few more sweeps when time allows. The reason I posted the first plot was how surprisingly even it looked. I've just been moving things around by ear before. |
bruno putzeys wrote on Sat, 04 September 2010 17:59 |
Having written my own measurement program from scratch I can tell you it's remarkably difficult to make a sweep-based (a la Farina) method that has no pre-ringing. I cracked it but it wasn't obvious. So you're quite likely to see something like this in a sweep based test whereas it wouldn't appear in an MLS based one. |
bruno putzeys wrote on Sat, 04 September 2010 20:16 |
On another note, the response measurement in the OP shows a very wide dip centered around 3kHz, and about 6dB deep. This worries me a bit because if this is actually there this is also going to affect your EQ'ing judgment. |
Quote: |
You should check with a gated measurement (just using the anechoic portion of a measurement done on one speaker) if this effect is just your room averaging out the power response or whether it's visible in the on-axis response of your monitors too |
Quote: |
If it is only the power response that's got this dip it's not as bad (though still worth being careful about when making EQ decisions) but if the on-axis response has the same dip you should consider correcting it. |
bruno putzeys wrote on Sat, 04 September 2010 20:16 |
The trouble is that the pre-ringing is at low frequencies, which is why it is that long. The sweep method uses an exponential sweep, which by necessity has to start somewhere (you can't start an exponential sweep at DC). The impulse you get after deconvolution is therefore high-pass filtered, and not necessarily minimum-phase. Hence the pre-ringing, unless you modify either the stimulus or the deconvolution kernel to add just the right amount of phase shift to make the high-pass portion minimum phase. |
bruno putzeys wrote on Sat, 04 September 2010 20:16 |
You should check with a gated measurement (just using the anechoic portion of a measurement done on one speaker) if this effect is just your room averaging out the power response or whether it's visible in the on-axis response of your monitors too. |
AndreasN wrote on Sun, 05 September 2010 13:28 |
The flat frequency response doesn't look particularly real either. My guess is that it's a response as measured across a portion of time that includes a lot of noise. The inclusion of the noise smooths the overall graph. |
bruno putzeys wrote on Sun, 05 September 2010 14:37 |
I know PMC and they are normally much flatter than this so it is worth putting some attention into finding what's going on. By correction I meant using an EQ but finding the root cause is going to be more effective. |
bruno putzeys wrote on Sun, 05 September 2010 14:37 |
The Y scale in the plots you refer to is stretched out wider. Anyhow, my main point is that measurement programs don't all do exactly the same thing. My sweep test doesn't have pre-ringing, and Farina's later paper mentions other means of reducing the problem. |
bruno putzeys wrote on Sun, 05 September 2010 14:37 |
Getting the anechoic response simply means chopping off everything starting from the first reflection. In a real room this means you get only a few ms to work with, so the measurement is only meaningful from a few 100Hz upward. But it would settle the question whether the anechoic on-axis response has a dip too. |