How to invert the frequency response of an IR file?

[pipelineaudio]

I want to make a corrective sweep so that when I send thru a signal path it returns as totally flat, so that I can really do an apples to apples comparison of things.

Any good ideas on how to do that?

[brainwreck]

Maybe explain more about what you are doing?

Unless I'm missing some info, it seems that what you want to do is remove the frequency response of a system. But I'm failing to see how that would be useful for doing any type of comparison.

[karbomusic]

Use the sound card calibration in REW if possible, that's what it's for - to help take everything but the device under test out of the equation.

[ashcat_lt]

How bout we try to answer the question before we get all philosophical.

If you’re doing a test sweep (not a step function impulse), then you must be using some process to deconvolve the sweep from the recording. Most of the time you tell the deconvolver where to look for the sweep file. If you point it at a file that includes the frequency response you want to remove, It should theoretically do what you’re asking. How well it actually works in practice I can’t say.

[karbomusic]

You can put anything you want calibrated out of the equation in the signal path, and it uses a sweep - Save that calibration file for later use - I'm not seeing a reason to reinvent the wheel here unless there is some reason you simply can't do it this way.

[Lokasenna]

Originally Posted by pipelineaudio

I want to make a corrective sweep so that when I send thru a signal path it returns as totally flat, so that I can really do an apples to apples comparison of things.

Any good ideas on how to do that?

Why not split the signal path prior to the chain you're testing so you have a dry copy, invert it, and then recombine them afterward?

Or you could keep them as separate channels and use SPAN's Difference mode.

[brainwreck]

Is the point here to separate the system from the signal for comparing the effects of systems? If so, yea, what Lokasenna said. Just flip the phase of the signal that hasn't passed through the system and sum, making sure that the two files are time aligned and level adjusted for the most cancellation. You could do that without physically splitting the signal.

I was just reading a bit about this last night over at beatdruid.com (Airwindows wiki). The idea being talked about was to separate what a processor is doing from the signal, for further processing the isolated effects of a process of interest to taste (not processing the output of a process working on a signal...serial), before putting them back together. It's an interesting idea, I thought.

I think this is the same thing that Pipelineaudio is getting at; isolating what some process is adding from the input.

[karbomusic]

I'm assuming pipe's trying to get his preamp myth testing done with charts and graphs so to speak based on his other thread. If so, it's really better to just use a test tool that takes non-DUT items out of the picture, especially if he is going to be generating response graphs etc. Maybe he's doing something else.

[brainwreck]

Originally Posted by karbomusic

I'm assuming pipe's trying to get his preamp myth testing done with charts and graphs so to speak based on his other thread. If so, it's really better to just use a test tool that takes non-DUT items out of the picture, especially if he is going to be generating response graphs etc. Maybe he's doing something else.

Yea, if it's only external hardware devices that will be tested, something like REW would probably be easiest to deal with. If it is internal dsp stuff, I don't know if a tool exists for that. I would love to have a jsfx tool for that.

[pipelineaudio]

Originally Posted by Lokasenna

Why not split the signal path prior to the chain you're testing so you have a dry copy, invert it, and then recombine them afterward?

Or you could keep them as separate channels and use SPAN's Difference mode.

How would I invert it? I think thats where I'm stuck

[pipelineaudio]

Hey thanks everyone! I have been sent a ton of hardware for these tests already and thank you to everyone who did that! I will be sending it all back ASAP

So thanks Karbo for pointing at REW. I did end up buying rightmark pro, but I am happy to use REW, I just couldn't get it to do what I wanted it to

I'll try to explain what I'm trying to do and maybe you guys have ideas:

The purpose, is to have something like my round trip latency chart, where there are actual, quantifiable numbers to at least show the measurable parameters

No more "warm", "sterile", "ephonic" "harmonius" or other Deepak Oprah Choprah crap.

A lot lot lot of the claims made can be tested directly such as when they claim magical frequency responses or noise levels.

I aim to test noise, distortion and frequency response, but theres a few tricks to this

I do have to have a decent methodology so that it is truly apples to apples

I would love to do like AP win where it does a level sweep to chart distortion, but I am actually ok to do say -20dBFS, -12 dBFS, -6 dBFS and -3dBFS and show for each

One of the concerns is that most testing software tries to do most or all tests at full scale, and then really Im just measuring converter distortion. I'd like not to do that, especially for the frequency responses...and if the methodology is always stable, then these things will be relative to each other, so the absolute noise level wont matter so much as compared between them

The converters are all flat enough (even the cheapest) so thats not that much of a problem, but in order to get down to mic level, I'm using DI's and Reamps and such, not 100% sure which yet, still testing for the lowest noise and distortion, and they all have a frequency response I'd like to flatten out for the test.

REW's callibration mode sounds great for that!

Karbo, I was really unable to figure how to get REW to work this way for me and how to set the different levels for the distortion tests. Also, Rightmark allows you to make charts that show up to 8 different devices at once which are cool, anything like that in REW?

Also, any way to just have one channel showing in REW? Or to have one channel plus a reference channel showing?

[karbomusic]

I thought it only showed a single channel and you had to pick a single channel for doing the measurements but could be wrong. It looks like you can overlay about anything but distortion:

[Lokasenna]

Originally Posted by pipelineaudio

How would I invert it? I think thats where I'm stuck

The phase button on a track - that's how I've always done null tests. If it was all within the same track or something there's also JS Channel Polarity Control.

[pipelineaudio]

But that won’t really invert the frequency response, it’ll invert all the differences, andnotmsure that’s what I want to do

[Lokasenna]

You want a perfectly neutral chain + the dry signal to = a flat line, yes? That's a null test. Anything above the line is signal that the chain added, anything below the line is signal that the chain cut.

[pipelineaudio]

You think there's a way to mod the test files to do this?

http://audio.rightmark.org/download.shtml

You can do "generate wav" to send out their test files, not sure if you could reimport modded ones AS the test files...Maybe

[pipelineaudio]

For REW distortion isnt part of the measurment right? You do that separately in the RTA?

[karbomusic]

Originally Posted by pipelineaudio

For REW distortion isnt part of the measurment right? You do that separately in the RTA?

No, it comes with the sweep. Everything in that particular window is from a measurement except for maybe scope - some of them (like waterfall) you have to click the generate button before it displays but distortion is there right after you measure IIRC. If I were to do it with REW, what I think I would do is create a calibration file using your chain minus the gear you are testing, save that file and use it as your calibration. As long as that file is loaded as a sound card calibration, the subsequent gear tests will have the effect of the chain subtracted from the gear response.

If you do it that way, then each time you click measure, you get a new entry over on the left that you can name to match the gear. Then you'll end up with one of those entries for every device you test, the only thing missing (I think) is being able to easily overlay them (depending on the graph you are using) some have overlay some don't. Once you have all of those, you can navigate between etc. and also save a project file that will keep the sweep results for all devices you saved with that session.

Note you can also change the calibration after the fact depending on how you do it - notice the "Change cal" button for each result in my screenshot above. Those were real tests from when I was comparing 3 sets of studio monitors I had at the time.

[pipelineaudio]

I'm getting the hang of it, but theres a few things

1: It seems like the spectrum analysis is smoothed out a lot, maybe not but it feels like it compared to RMAA

2. For the spectrum window that comes up after the measurement ("All SPL") is there anyway to set the scale to dBFS and have 0 at the top? Alternatively, could I do frequency sweeps in the other spectrum window, the real time one? The nice thing about RMAA is that I can compare very simply, different devices...seems like the functionality is here on REW, I just cant quite figure it out yet

3. I'd rather see this type of distortion plot, or in addition I guess

[karbomusic]

At the top right of the All SPL graph, click the "Controls" button, there's an offset you can enter for each measurement but not sure if that's what you really need. It will at least allow compensating for level changes. I think there's a smoothing control there as well but mine was off by default. I was able to create an audio chain calibration file, then use that as a calibration against some test measurements I was just playing with FWIW. Looks like you can create as many types of calibration footprints as you need (.cal), then apply those as needed to measurements - if you set it as default sc calibration in settings, it will automatically be attached to each measurement.

[karbomusic]

Reg #3: Isn't the plot above measuring a 1k sine instead of a sweep?

[karbomusic]

If you want to use a sine test signal such as 1k, you can use the generator to play the signal, then the RTA to record it and the THD. Just sort of figuring it out as I go. If I'm reading this right, the loudest harmonic below (3rd) is about -45.7 dbFS (-43dBFS + -2.7dBFS peak). The THD is 1.44% but I had to clip the A/D some to get something to show up. If you hover your mouse to the left of that little overlay with the stats/THD numbers, a little box pops up allowing you to set the scale to dBFS instead of dB, aka zero on the top.

[karbomusic]

After playing with it for the last hour, it seems that the most critical piece is keeping the levels for each measurement exactly the same if possible. Some graphs can be offset per measurement result, but seems to me if they are captured methodically and at the same level, everything downstream gets a lot easier. It would be nice however if you could add a per dB offset per measurement that was global to all graphs.

One other problem I see is if the gear, such as a preamp without an output control, you won't be able to push it hard enough to test it when pushed without clipping the A/D.

[pipelineaudio]

Originally Posted by karbomusic

Reg #3: Isn't the plot above measuring a 1k sine instead of a sweep?

Yup. That’s one of the tests I want

[pipelineaudio]

Originally Posted by karbomusic

One other problem I see is if the gear, such as a preamp without an output control, you won't be able to push it hard enough to test it when pushed without clipping the A/D.

That would be the same issue for the hyperbolic claims anyway so it’s ok. I think I may be able to get rew to do what I want

[karbomusic]

Originally Posted by pipelineaudio

That would be the same issue for the hyperbolic claims anyway so it’s ok.

Some do have output controls and/or may use other external attenuation. If a device you are testing has an output as you are testing, seems like it should be tested both as normal and a little saturated - unless the test specifies nominal levels only.

[pipelineaudio]

I just cannot figure out a way to make the one that can show multiple traces start with 0dBFS at the top

[pipelineaudio]

Man I am stuck with this stuff...I've sent emails off ot the companies involved but nothing back yet...is there a forum community someplace for RMAA and REW?

[karbomusic]

REW has a forum, probably can search for it...

https://www.avnirvana.com/forums/off...port-forum.10/

My finding the other night was one of the SPL or RTA windows allowed setting dbFS which places 0dB at the top however...

I came to the conclusion that it wasn't easily possible to have that and multiple results overlaid. That said, I'd probably it's builtin capture screenshot feature and just overlay them in photoshop or something but that's stuff I already have and know how to do <- but that would have taken less time to do than we spent searching on doing it natively thus far.

[pipelineaudio]

I left a post at the forum...fingers crossed

[pipelineaudio]

Any idea how to test noise

and especially, to get rid of this stuff

[pipelineaudio]

I'm making some real progress today! Hopefully I'll have something soon

[karbomusic]

[pipelineaudio]

Starting to narrow down the methodology! Any idea why on the distortion tests, the Focusrite Octopre would look so different than all the rest of these? Does the same sort of thing on REW

[karbomusic]

Maybe it sucks. - Actually I have no idea, either it has more distortion or something was overlooked.

[pipelineaudio]

jeesus....coming up for air for a second....dont ever get into this, make someone else do it!

[brainwreck]

Ha ha. Sounds like good times over there.