[bigtoga]

If I use a length of 128, my graphs look reallllllly nice and impressive (to me anyway). The higher up I go in length, the worse they look. At 512k, I get a complete null drop at every 50: 50hz, 150hz, 250hz (see attached).

I'm thinking that my PC must only be able to handle the 128 Length.

The docs seem to imply that I don't have a fast enough CPU or enough memory to do 512 or 1024:

Longer sweeps provide higher signal-to-noise ratio (S/N) in the measurements, each doubling of the sweep length improves S/N by almost 3dB. However, the time required to perform the processing after each sweep will more than double for each doubling of sweep length. If REW is running on a computer which does not have a fast processor and a lot of memory, measurements will be much faster using the shortest sweep length (128k samples), at a small S/N penalty of about 3dB compared to the default. 1GB of RAM and a fast processor are recommended if using the 1M sweep, invalid measurements may occur on computers which have insufficient RAM or processor speed

The weird thing is that I have 3.5GB RAM and a fast AMD quad core CPU when I run the tests yet clearly 512 is "too much" for my system.

So my two questions would be:

• What am I giving up if I only go with 128 (instead of 256)?
• How can I tell whether 256 is wrong/messed up/invalid measurement?

When I do 512 or 1024, it's obvious that I have an invalid measure because of the dips at the 50s. But at 256, I have dips but they aren't completely null like they are at 512. But... if I measure same speaker/distance/test but use 128, there aren't any of those nulls.

Check out the 128 and 256 graphs - at around 580hz, I have a dip. In the 128 measurement, it's about a 7db dip but in the 256 measurement, its over 30.

Which is "right" and how do I know when I'm working with a graphing Level that is invalid?

 

[JohnM]

There is a feedback path somewhere that has a 40ms delay in it when you use the longest measurements, hence the nulls at frequencies that correspond to odd multiples of 50Hz (the lowest frequency where 40ms gives a 180 degree phase shift is 50Hz). It might be some buffering issue somewhere, difficult to say.

The 128k and 256k both look OK, at 128k you lose a little in S/N (about 3dB vs 256k) and a little in frequency resolution (which is why you see the depth of the notch better at 256k).

 

[bigtoga]

There is a feedback path somewhere that has a 40ms delay in it when you use the longest measurements, hence the nulls at frequencies that correspond to odd multiples of 50Hz (the lowest frequency where 40ms gives a 180 degree phase shift is 50Hz). It might be some buffering issue somewhere, difficult to say.

The 128k and 256k both look OK, at 128k you lose a little in S/N (about 3dB vs 256k) and a little in frequency resolution (which is why you see the depth of the notch better at 256k).

Good to know - thank you. I tried all the various buffers (16-128) but that offered no change when I was 512 or 1024.

Can I ask a sort of related question? When I'm looking at my room for evaluation, what's the smoothing that I should be looking at? When I look at 1/3 octave, for example, it looks really nice lol. Here though there is a HUGE -30db dip around 580. When I put on 1/3 octave smoothing, it works out to be about a 3db dip. This particular room has about 25 panels of OC703 and OC705 in it (12'x13'x8'). If I need more, I'm fine adding it in - just looking for guidance

Thanks again, John, for all of the help!

 

[JohnM]

At frequencies above about 200Hz 1/6 or 1/3 octave smoothing is sensible, those narrow notches you see at higher frequencies vary dramatically with position and have little relevance to what we perceive.

 

[bigtoga]

Brilliant - thank you. I saw what appeared to be crazy comb filtering about 8k when I removed the smoothing but it was fine when I went to 1/3.

:T

 

[bigtoga]

Oh - I have "follow ups to the follow up" question if that's allowed!

Question 1: Given what you said, it would seem logical to me that, as frequencies rise, so should your smoothing. This is what makes sense to me:

• 0hz - 200hz: No smoothing
• 200hz-1000hz: 1/3 octave
• 1000 - 4000: 1/6 octave
• 4000-12000: 1/12
• 12000-16000: 1/24th
• 16000-20000: 1/48th

And I just halfway-randomly picked those values - just thinking that, as I move closer to 20k, I would want more and more smoothing. Is that correct?

Question 2: When I look at the graph above with no smoothing (the 256 one), when it shows the massive dip at 570hz, since that's above 200, should I ignore it and go with 1/3 octave? The 1/3 octave still shows a dip but it is more along the lines of a 15db dip vs. the 30db shown here.

Thank you!

 

[JohnM]

You actually get more smoothing the lower the octave fraction number is. The faction is what part of an octave the filter spans, so a 1/3 octave filter covers a wider span than 1/48 and smooths the result more. There is certainly value in variable smoothing though, and it is on the list of things to add.

You can ignore dips at any specific frequency above the bass range, but if the response overall looks very uneven despite smoothing then additional treatment could help to smooth it. The folk in the Home Audio Acoustics forum can offer more advice on that.