Home Theater Forum and Systems banner
1 - 10 of 10 Posts

·
Registered
Joined
·
60 Posts
Discussion Starter · #1 ·
John, why is it that when I have a frequency response that I want to EQ and I hit the Match Response To Target button, if I then hit Reset Filters for Current Measurement (to erase the suggested filters and start again) and then again hit the Match Response to Target again, a different set of filters always comes up. I can do this over and over again and the filters suggested are not the same. I am leaving all the other Filter tasks at their defaults. Now overall each predicated response from the suggested filters give an overall similar response, i.e. it closely matches the Target Line but sometimes for a response the program can suggest 4 filters and then Resetting and getting another set of filters suggested can be result in 11 filters the next time around even if all the other parameters like I said before stay the same. Since I want to try and get the minimum number of filters to be used I will often repeated hit the Match Response and the Reset Filters until I get a smaller number of filters but when I get what might be a smaller number of filters (say 3,4,or 5) I have to remember to save them or else quite often I cannot get the program to suggest the smaller number of filters again.
 

·
Premium Member
Joined
·
108 Posts
I know you said you're keeping all the Filter Tasks at their defaults, but how about the Target Settings? The way you describe it looks like you have different values for the Target Level...

I get the same filters every time I run the EQ...
 

·
Registered
Joined
·
678 Posts
John, why is it that when ... a different set of filters always comes up. ...
I asked a similar question in the V5 Beta thread, and learned from John that the equalizer is a "stochastic optimiser using random perturbations, so results can vary from one run to another."

I understand this to mean that, because of the nature of the goal being an equalization that is "close enough" to the target, the stopping point is dependent on the starting point. By varying the starting point, i.e., performing multiple runs, you can choose to keep the one that you like best.

Bill
 

·
Registered
Joined
·
60 Posts
Discussion Starter · #4 ·
I know you said you're keeping all the Filter Tasks at their defaults, but how about the Target Settings? The way you describe it looks like you have different values for the Target Level...

I get the same filters every time I run the EQ...
No, I have the same Target Level everytime. Maybe it depends on how varied/complicated your initial response is that you are trying to equalize.
 

·
Registered
Joined
·
60 Posts
Discussion Starter · #5 ·
I asked a similar question in the V5 Beta thread, and learned from John that the equalizer is a "stochastic optimiser using random perturbations, so results can vary from one run to another."

Bill
I guess that may explain things but like I said, with the current response in my current room, I can get a suggestion of 5 filters one time up to 11 filters another time all with the same initial measurement and no other settings altered. In all cases, the predicated response looks very similiar in that it follows the Target line but I will take the smaller number of filters to get a similar result anytime.
 

·
Elite Shackster
Joined
·
7,142 Posts
I have a similar question if you dont mind me jumping in. How come it sometimes sets several filters almost identically? I could understand same centred frequencies with different values, but Ive seen the auto target set 3,4 or 5 filters identically which doesnt make sense to me.
 

·
Registered
Joined
·
60 Posts
Discussion Starter · #7 ·
I have a similar question if you dont mind me jumping in. How come it sometimes sets several filters almost identically? I could understand same centred frequencies with different values, but Ive seen the auto target set 3,4 or 5 filters identically which doesnt make sense to me.
With my responses, I have never seen 3,4 or 5 identical filters stacked on top of one another before. I know there is an option to disable or enable the use of doubled filters so I can see maybe the use of 2 identical ones but 5?
 

·
Elite Shackster
Joined
·
7,142 Posts
I ran some filters on an IB response and it cam up with maybe 5 overall vastly different filters out of the entire set, some of the filter were stacked a lot, which I was rather surprised at. The only thing I can think of is if the full amount of filters simply werent required and REW was trying to effectively get rid of some. That didnt seem to add up with the response on show though, as I felt it could have been improved had the filters not been stacked so much, but thats a guess really.
 

·
Premium Member
Joined
·
6,711 Posts
John, why is it that when I have a frequency response that I want to EQ and I hit the Match Response To Target button, if I then hit Reset Filters for Current Measurement (to erase the suggested filters and start again) and then again hit the Match Response to Target again, a different set of filters always comes up. I can do this over and over again and the filters suggested are not the same. I am leaving all the other Filter tasks at their defaults. Now overall each predicated response from the suggested filters give an overall similar response, i.e. it closely matches the Target Line but sometimes for a response the program can suggest 4 filters and then Resetting and getting another set of filters suggested can be result in 11 filters the next time around even if all the other parameters like I said before stay the same. Since I want to try and get the minimum number of filters to be used I will often repeated hit the Match Response and the Reset Filters until I get a smaller number of filters but when I get what might be a smaller number of filters (say 3,4,or 5) I have to remember to save them or else quite often I cannot get the program to suggest the smaller number of filters again.
The long winded answer is that correcting a response is a multi-variable optimisation problem that can easily get stuck in local minima and so fail to find a global optimum. Using a random perturbation approach greatly increases the chances of finding a good global optimum result, but the aim of the optimisation, to get a close match to the target, can be satisfied to the same degree in many different ways, for example being 1dB away from the target at 50Hz and exactly on it at 100Hz is as good as being exactly on target at 50 and 1dB off at 100 from a matching perspective. As a result there can be small differences in the results of different runs, though the main features (the corrections for big peaks) should be quite stable - the variation tends to be in the smaller corrections. If you want REW to use fewer filters only enable as many as you want to use and that is how many REW will apply, doing the best it can with the number of filters you allow.

On the observation that several filters sometimes result that are very close to one another, that can occur if there is a deep notch or a very sharp transition in the response and boost filters are allowed. If anyone has specific examples of measurements that exhibit this do send them to me along with the filter optimisation settings you were using so that I can use them to tune the algorithm.
 

·
Elite Shackster
Joined
·
7,142 Posts
On the observation that several filters sometimes result that are very close to one another, that can occur if there is a deep notch or a very sharp transition in the response and boost filters are allowed. If anyone has specific examples of measurements that exhibit this do send them to me along with the filter optimisation settings you were using so that I can use them to tune the algorithm.
The file I had at the time would have been ideal for that, but its not going to be possible to send it now. I will keep that one in mind for the future though.
 
1 - 10 of 10 Posts
Top