What is the most meaningful way of measuring frequency response? L+R, average of the two channels, or something else?

WillDoLikeMildew

New Member
Thread Starter
Joined
Jun 27, 2024
Messages
10
More  
Preamp, Processor or Receiver
Yamaha RX-V385
Front Speakers
Dali Zensor 1
Center Channel Speaker
none
Surround Speakers
Cambridge Audio Minx Min 22
Subwoofers
SVS SB-1000 Pro
Other Speakers
miniDSP 2x4 HD
I recently learned that when measuring the left and right channel playing simultaneously, differences of mic placement of only a couple millimeters could result in large differences in the measured level of top end, even with 1/1 smoothing applied, due to the phase discrepancies between the speakers relative to the mic position. This introduces an arbitrary amount of error that won't be subjectively experienced by a person with two ears who doesn't sit frozen in the exact same millimeter of space every time they listen to music.

So now I'm wondering, what is the actual most meaningful way to measure the frequency response of an entire stereo system? I see people sharing their measured frequency responses all the time, so what are they doing? Are they just ignoring this potential error?

Are they taking an average of the measurements of the left and right channel instead? The issue I see with doing that is in a system with one subwoofer (like my own system), muting one of the channels will make the sub experience a significantly larger dB decrease than the speakers. This is because the channels feeding the sub are 100% perfectly in phase with each other, whereas the channels for the speakers won't be once they enter the real world and come out of two entirely different locations.

It seems then that a measurement of the left and right channels playing together would be more accurate for the low frequencies, whereas an average of the left and right channel measurements would be more accurate for the high frequencies.

So what I do is I get both the measurement of L+R playing simultaneously, and an average of L & R playing alone. The average is obviously going to be quieter than the sum of L+R. So I apply some gain to the average to make it reach the same level as L+R, using the midrange as a reference (around 500-1000 Hz), then I just stitch together the two curves with L+R being used from 20-1000 Hz roughly, and the average of L&R being used from 1000Hz - 20kHz.

Am I overthinking this? What are people actually doing when they share their frequency responses? Is there an easier way to make sure I'm getting an accurate, meaningful read of my frequency response from a listening position?
 

sm52

Member
Joined
Mar 14, 2019
Messages
1,001
The correct way is to measure each channel separately. And the subwoofer separately. From one point in space, for example from the center of the listening position. Using acoustic synchronization. You can then use REW math to generate collaborative responses. To obtain the correct sum of the right and left channels, one of them must be aligned in time so that their sum at the end of the HF gives a smooth addition without waves and subtractions.
 

WillDoLikeMildew

New Member
Thread Starter
Joined
Jun 27, 2024
Messages
10
More  
Preamp, Processor or Receiver
Yamaha RX-V385
Front Speakers
Dali Zensor 1
Center Channel Speaker
none
Surround Speakers
Cambridge Audio Minx Min 22
Subwoofers
SVS SB-1000 Pro
Other Speakers
miniDSP 2x4 HD
The correct way is to measure each channel separately. And the subwoofer separately. From one point in space, for example from the center of the listening position. Using acoustic synchronization. You can then use REW math to generate collaborative responses. To obtain the correct sum of the right and left channels, one of them must be aligned in time so that their sum at the end of the HF gives a smooth addition without waves and subtractions.
Ok, so if I'm understanding this correctly, I measure L (without the sub), R (without the sub), and the sub, all using acoustic synchronization. And then I use the "alignment tool" to sum L and R with whatever delay makes their top end the most in phase. And then I sum that result with the sub.

Is that correct? That makes a lot of sense and is way simpler than whatever the hell I was doing.
 
Last edited:

sm52

Member
Joined
Mar 14, 2019
Messages
1,001
Yes.

To be completely correct, you need to change the sub's time (its t=0) to the value that was obtained by adding L and R. If L had to be done earlier, then the microphone has 'moved' to the left. If the sub is on the left, then it must be done earlier by the same amount. But if you place a microphone approximately halfway between L and R, then the offset will be very small and will change almost nothing for the sub. So you need to adjust the subwoofer if you have set yourself the task of matching these three speakers.
 
Last edited:

WillDoLikeMildew

New Member
Thread Starter
Joined
Jun 27, 2024
Messages
10
More  
Preamp, Processor or Receiver
Yamaha RX-V385
Front Speakers
Dali Zensor 1
Center Channel Speaker
none
Surround Speakers
Cambridge Audio Minx Min 22
Subwoofers
SVS SB-1000 Pro
Other Speakers
miniDSP 2x4 HD
Is there a way to choose delay times with more precision than 0.01 ms? Adjusting by this amount within the alignment tool actually creates a pretty significant change. I noticed that the calculated delay when using acoustic synchronization is much more precise than this and I bet if I was able to enter the difference between the calculated delay of L and R on the alignment tool, it would be right on the money.
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,930
0.01 ms corresponds to about 3 mm change in position. How still are you planning to sit?
 

sm52

Member
Joined
Mar 14, 2019
Messages
1,001
Is there a way to choose delay times with more precision than 0.01 ms?
Yes. Load two measurements into REW, Overlays, Impulse. One of the latest updates introduced the Time align function, which can be seen by right-clicking when the cursor is on the graph field. This is a quick way to align the time of two measurements. There is also Offset t=0 with an accuracy of 1 μs. You can move one graph, then click Apply & next, and move another graph. But at the same time you do not see the overall frequency response. Therefore, you need to know the exact delay figure. Also, Offset t=0 can be set with even greater accuracy for one measurement in the main window, Impulse tab.

John, I apologize for interfering.
 

WillDoLikeMildew

New Member
Thread Starter
Joined
Jun 27, 2024
Messages
10
More  
Preamp, Processor or Receiver
Yamaha RX-V385
Front Speakers
Dali Zensor 1
Center Channel Speaker
none
Surround Speakers
Cambridge Audio Minx Min 22
Subwoofers
SVS SB-1000 Pro
Other Speakers
miniDSP 2x4 HD
Yes. Load two measurements into REW, Overlays, Impulse. One of the latest updates introduced the Time align function, which can be seen by right-clicking when the cursor is on the graph field. This is a quick way to align the time of two measurements. There is also Offset t=0 with an accuracy of 1 μs. You can move one graph, then click Apply & next, and move another graph. But at the same time you do not see the overall frequency response. Therefore, you need to know the exact delay figure. Also, Offset t=0 can be set with even greater accuracy for one measurement in the main window, Impulse tab.

John, I apologize for interfering.
Thanks. Adjusting t=0 in the impulse window before combining with the alignment tool works. This level of precision didn't actually end up changing the combined frequency response by anything significant for me but this is good to know.
0.01 ms corresponds to about 3 mm change in position. How still are you planning to sit?
It's more so that I'm trying to achieve a consistent way of measuring the frequency response of the system that is repeatable and doesn't introduce much error. The position of the mic can effect the level of top end by as much as a few decibels, which doesn't bother me in the real world, but if I were to go into my receiver and change the top end by a few decibels I would still notice that difference even if my head is in many different listening positions. The point in space at which the speakers are time aligned is just a well-defined reference point so that if I repeated this process again and things were different I would know it wouldn't have to do with time alignment. In my case, I am planning on moving soon and when I set up my gear in my new place I would like to create a similar frequency response that I have now, and as long as my measurements here and my measurements in the new place are time aligned in the exact same way I would know the comparison between the two is meaningful.

That being said, when I used more precision with the t=0 method, my result was barely different at all. But maybe this won't always be the case.
 
Top Bottom