advertisement


measurement microphone calibration

Could you elaborate a little on this, @h.g. ?

If people want to know more I would suggest quietly reading something like the B&K literature I referred to above or equivalents.

When you say the measurements take a long time to obtain are you referring to a stepped sine measurement? REW's default automated sine sweep takes only 6 seconds to complete and can be viewed without smoothing, and REW's RTA function can be displayed with 1/48th smoothing which shows almost as much detail at low frequencies as the unsmoothed sine sweep.

I wasn't referring to any particular type of measurement. To adequately resolve a low frequency high Q resonance you will need to use a large number of samples in the FFT. A single raw measurement will contain a significant amount of grass/wiggles/noise. Smoothing/averaging in frequency will reduce this but it will also smooth a high Q resonance into a lower Q one. If you ensemble average (average a sequence of independent measurements) the frequency resolution will be fully preserved and the grass/wiggles/noise will reduce but only slowly with increasing numbers of measurements. I don't know if REW can show the evolving averaged frequency spectra from playing something like white noise through the speakers but this would show the process nicely in real time as the wiggly spectra slowly becomes smoother as time evolves.

With regards to the number of measurements, is there not a risk that taking too many measurements spread out around the MLP will essentially mask the specific frequency and amplitude of the resonances heard at the MLP?

Google give MLP as My Little Pony. I am guessing LP is listening position but I am not what you consider might be changing over time.
 
If people want to know more I would suggest quietly reading something like the B&K literature I referred to above or equivalents.



I wasn't referring to any particular type of measurement. To adequately resolve a low frequency high Q resonance you will need to use a large number of samples in the FFT. A single raw measurement will contain a significant amount of grass/wiggles/noise. Smoothing/averaging in frequency will reduce this but it will also smooth a high Q resonance into a lower Q one. If you ensemble average (average a sequence of independent measurements) the frequency resolution will be fully preserved and the grass/wiggles/noise will reduce but only slowly with increasing numbers of measurements. I don't know if REW can show the evolving averaged frequency spectra from playing something like white noise through the speakers but this would show the process nicely in real time as the wiggly spectra slowly becomes smoother as time evolves.



Google give MLP as My Little Pony. I am guessing LP is listening position but I am not what you consider might be changing over time.

I think the different language used may be the source of the confusion, that said I've still to read the literature you suggest so could have picked up the wrong end of the stick entirely!

The RTA in REW allows you to set the FFT as high as 131072 but I stick to the default 65536. You can also set it to show a rolling average of the previous n samples (n can be set anywhere between 2 and Forever). I use the RTA to make real time adjustments to my EQ so that I can see its effect, but a sine wave sweep is the only way REW can analyse phase, impulse, distortion, decay times, etc.

The default sine wave sweep is a 256k log sweep that lasts 5.9 seconds. This can be increased to a 1 million log sweep lasting 23.8 seconds but I'm not sure what benefit increasing the resolution brings given that the Q of the modal peaks displayed in my measurements is already narrower than what my parametric EQ software is capable of (the narrowest filter I can apply is Q=20).

BTW - MLP is an abbreviation of Main Listening Position. I too had to ask for a definition when I first encountered it but I assumed you would've been familiar with the term.
 
With regards to the number of measurements, is there not a risk that taking too many measurements spread out around the MLP will essentially mask the specific frequency and amplitude of the resonances heard at the MLP?

Since I'm mostly looking at this as a [ultra-amateur, bedroom-bound] studio project, there is a single listening position. I believe @h.g. also was referring to a studio environment, albeit with obvious professional application.

With a single listening position, and thus keeping the mic in a fixed position, any variation in resonance at a given frequency should regress to the mean over an infinite number of samples. As you collect more independent samples (i.e. record more sweeps) and average them across the full frequency range, your estimate of the mean SPL at a given frequency will become more and more accurate. Just how accurate depends on the variance of each sample and the number of samples, which you would calculate as the standard error of the mean: the sample standard deviation (square root of the variance) divided by the square root of the number of samples. This would be measured in dB SPL just as the actual mean response is, so you would know that, e.g., at frequency X Hz, your mean response is Y dB +/- 0.5dB or whatever.

My intuition says that in a small room at lower frequencies the variance should be lower than at higher frequencies in most cases (can someone confirm?), so extrapolating from that hypothesis, you should need fewer measurements to dial in the mean SPL at a problematic low frequency than you would for a higher frequency. There is no fixed number of samples though and it depends on just how much variance you're seeing and just how accurate of an estimate you want.

Edit to say: all of that was a guess! I have no depth of experience. Maybe @h.g. did mean averaging over samples read at different positions.
 
OK an update. I've taken a few days off work so I took advantage of the opportunity to do some more measurements. I tried to be a bit cleaner and more systematic in my approach. This time, first of all, I used a proper stand for the mic to reduce any possible complications from other reflections from my previous creative use of a table on its side (although they were probably minor to begin with). Also, this time I positioned the mic in a vertical position. It's hard to find a solid recommendation online (hopefully more info will come from some book learnin'), but it seems that for my purposes that would be the better orientation. I opted to use multiple mic positions, but given the nearfield, single listening position arrangement, these were fairly minor in their variation. Basically, I tested 6 positions: forwards/backwards x left, center, right. These were measured separately for the left and right channels with 2 repetitions per position. The final curves used to generate the filter impulse files, then, were the averages of 12 curves for each channel. I would have liked a few more reps but my wife is working downstairs and I cannot imagine that all those filter sweeps were very enjoyable to hear....

Before measuring, I tweaked the filter settings on my monitors, increasing the LF shelf attenuation to tame the bass further. For constructing the EQ, I more carefully tuned the target curve, which included setting a more representative high frequency roll-off (my monitors start to roll off around 4000Hz, whereas the default in REW is 1000Hz). This allowed me to set the target level with more confidence. I allowed a certain amount of boosting, since it seems the 600-800Hz range was attenuated compared to the rest of the curve. Unfortunately, for the right channel there is a significant dip around 100Hz that I cannot fix and that would probably need room treatment.

Anyway, long story short, after dialing things in with much more care and attention, I'm very happy with the result. The very noticeable peak around 150Hz has been cleanly cut out, however unlike my previous attempt, the result does not sound anemic in the low-end. In fact it sounds very full-bodied (lack of sub-bass due to no subwoofer aside...).

So, I'll rest easy about the calibration of the mic for now (the original impetus for this thread). I also feel a bit more confident in the overall procedure, however I'm going to do some reading to better my understanding. In some ways this was a trial run, since I will be moving all of my stuff to a different room later this year, so I'll have to do it all over again. The advantage is in that room I can probably get away with at least a bit of room treatment....

Thanks for all the advice, everyone.
 
Since I'm mostly looking at this as a [ultra-amateur, bedroom-bound] studio project, there is a single listening position. I believe @h.g. also was referring to a studio environment, albeit with obvious professional application.

With a single listening position, and thus keeping the mic in a fixed position, any variation in resonance at a given frequency should regress to the mean over an infinite number of samples. As you collect more independent samples (i.e. record more sweeps) and average them across the full frequency range, your estimate of the mean SPL at a given frequency will become more and more accurate. Just how accurate depends on the variance of each sample and the number of samples, which you would calculate as the standard error of the mean: the sample standard deviation (square root of the variance) divided by the square root of the number of samples. This would be measured in dB SPL just as the actual mean response is, so you would know that, e.g., at frequency X Hz, your mean response is Y dB +/- 0.5dB or whatever.

My intuition says that in a small room at lower frequencies the variance should be lower than at higher frequencies in most cases (can someone confirm?), so extrapolating from that hypothesis, you should need fewer measurements to dial in the mean SPL at a problematic low frequency than you would for a higher frequency. There is no fixed number of samples though and it depends on just how much variance you're seeing and just how accurate of an estimate you want.

Edit to say: all of that was a guess! I have no depth of experience. Maybe @h.g. did mean averaging over samples read at different positions.
From a statistical perspective, yes, increasing the number of measurements taken at the fixed listening position should reduce variation, but in practice I do not find it to be that necessary, at least not for frequencies that are higher than my lowest room mode. It is however important to take the measurements when external noise sources (e.g. traffic, wind, etc) are at their minimum, and to take the measurements at an SPL that is sufficiently higher than the background noise floor.

If you live in a place where external noise is a problem, it's a good idea to check the Distortion profile of every measurement you take. If I see something odd in the Distortion profile, I repeat the measurement a few times until the Distortion profile looks normal.

As long as my measurements aren't corrupted by external noise, I observe essentially no difference in the amplitude of my low frequency measurements, or at least not enough that would cause me to significantly undershoot or overshoot the amount of EQ I apply to flatten a room mode. The +10dB peak you appear to have at 150Hz, for example, are you really going hear the difference between flattening it by a fraction of a dB too little or too much?
 


advertisement


Back
Top