I've discovered something velly velly intellesting about the EQ on my player.

For a while I thought I was hearing something wrong with my speakers or amplification... Some of you might recall that in another thread I recently stated that after I did my RTA adjustments that I couldn't get the system to go as loud as I could before (without distortion), but that it sounded better overall.

I think I've discovered that the distortion I perceived is actually digital clipping happening internally at the DSP stage, before the sound even heads out of the outputs.

Yet I've done nothing but cut frequencies in the EQ, so everything is below 0db and in fact the overall output of this EQ setting is much quieter than a flat EQ.

Attached here is an MP3 which demonstrates this clipping on my player. It's a snippet of "Witness" from the Sarah McLachlan album "Surfacing". On my player, each of the bass notes sounds fine with a flat EQ, but with my "tweaked" EQ, there is distinct clipping on each bass note when the volume is up near 0db. When I back the volume down to -7db or so, the clipping goes away. This is on any set of speakers or even a set of headphones plugged directly into the player, it doesn't have to be loud or outside of the handling capacity of the speakers or headphones.

Can anyone reproduce this problem for me and verify that I'm not crazy? My player settings are as follows to reproduce the distortion:

Hijack settings:
Force DC/Car mode (so I can get to the equalizer preset in question).
Auto Volume Adjust: Off
Bass Adjust: Off
Treble Adjust: Off
Left/Right Time Alignment: Off

Player settings in Car mode:
Volume: 0db
Loudness: Off
Balance: Center
Fader: Center

Equalizer Preset:
Mode: 10 band Locked L/R
(NOTE: Beta11 has a bug adjusting the Q in this mode. I'm running beta12 which doesn't have this bug. If you try to reproduce this in beta11, you have to put it in "independent" mode and do separate adjustments for Left and Right but just set them to the same values.)
Band 1: -16db 50hz Q5.00
Band 2: -10db 289hz Q2.91
Band 3: -5db 180hz Q5.00
Band 4: -8db 265hz Q19.6
Band 5: -24db 1.00khz Q7.17
Band 6: -9db 900hz Q0.67
Band 7: -8db 100hz Q3.00
Band 8: -10db 3.78khz Q0.84
Band 9: +0db 10.0khz Q5.00 (left at 0 so Genixia's bass/treble code can use it)
Band 10: +0db 18.0khz Q5.00 (left at 0 so Genixia's bass/treble code can use it)

Note that the above EQ settings aren't in a strict order of frequency (low to high) since that's the beauty or a parametric equalizer, a band is a band, they don't have to be in order.

Can someone else reproduce the bass clipping in this case? Or does anyone have a clue WHY this is happening? If you'll notice, every single one of my frequencies is CUT not BOOSTED, so in theory it should be REDUCING the chance for clipping instead of increasing it.

Two things will remove the clipping: Turning the player volume down several notches, or switching to a flat EQ (which is much louder overall).

What is it about the DSP's equalizer that doesn't like my settings here?

Umm... Tony? where did you get beta 12?

I'm on the alpha team. There are only a few of us on the BBS who are on the alpha team. And before you ask, it's by invitation only.

He's annoyingly one of the priviledged few that are in the tester group for empeg The rest of us have to make do with beta 11 for the moment

- Trevor

There's really little difference between beta12 and beta11 except for some bug fixes. The next public release will simply be more bug fixes.

So back on topic: does anyone have any idea why the DSP is doing this?

Busted!

How are you getting the feeling it is happening at the DSP and not the output? Are you tapping the audio stream as it is leaving the DSP? (Note: I didn't even know if this was possible, or the internal processing order. Just curious.)

Can you reproduce this in the home settings by copying the exact configuration to a home EQ setting with a standard kernel? This would eliminate the fader and the hijack code.

Lastly, is there a way you can put a scope on the output to see the signal? Worst case, record in through your soundcard and view the waveform with ProTools, Sound Forge, Goldenwave or equivalent. Just a visual verification that you are seeing clipping.

Additional: Also, can you verify what frequencies the clipping occurs on by using test tones? Is it across the whole spectrum?

Yeah but it's just newer and better (well more bugfixes) so we want it anyway

Do you get clipping no matter what equaliser settings you use apart from the flat one?

It could be possible that there is a mistake in one of the values for the DSP that are calculated when you set the equaliser. That could cause it to have a slight gain.

- Trevor

Do you get clipping no matter what equaliser settings you use apart from the flat one?

I have other EQ presets which do not clip. They are not as extreme as this one in terms of detailed Q adjustments etc.

It could be possible that there is a mistake in one of the values for the DSP that are calculated when you set the equaliser. That could cause it to have a slight gain.

That's what I'm wondering about.

Can anyone else reproduce this with the settings I supplied and the file I supplied?

How are you getting the feeling it is happening at the DSP and not the output?

Technically, I don't know. And I'm not sure that distinction needs to be made, even. Whether it's happening digitally inside the DSP, or analog in the line-level output stage just past the DSP, either way it's clipping when I haven't done anything that should have made it clip.

I am certain that the clipping is not happening in my amplification, speakers, or headphones. It's definitely happening somewhere internal to the player, I went to great pains to be sure.

Can you reproduce this in the home settings by copying the exact configuration to a home EQ setting with a standard kernel?

Haven't gone that far yet. I really hate re-installing the player software because then I have to re-do the replaced graphics for the visuals and also re-install XML interface.

Wait, Mark publishes a raw compiled Beta11 kernel doesn't he? Okay, I'll try that. Should only take a couple minutes to punch everything in...

Hmm. I can reproduce it with a plain kernel and while in "home" mode, by plugging those values into an unused preset. And I can hear it just as clearly.

However when I record it to a WAV file through the PC's Line input and view the wave in CoolEdit, I can't actually see any clipped peaks. I can clearly HEAR what sounds like clipping distortion even when previewing the WAV in CoolEdit, but I can't SEE it in the waveform.

And now suddenly I'm not so sure that it's in the player at all. Although it distorts on every other piece of equipment I've got (car system regardless of amp gain setting, home system regardless of volume level, computer line input regardless of input level, more than one pair of headphones including ones that have never distorted on any material before and which were plugged straight into the player), there was one case where it didn't distort.

That was via the MICROPHONE input on my PC. It didn't distort on the microphone input when I turned off the gain boost. And I can crank the PC's volume up really really loud while it's using the mic input, and it sounds GREAT. Whereas, with the line input (and every other possible audio output including my car system), it distorts even at very low volumes or amp gain settings.

But you're not supposed to plug line-outs into a Mic input. It's a different impedance. Odd.

What is it about the PC's mic input that would make it change the way the player's output stage behaves in this situation? Did the mic input place a higher "load" on the line outputs or something?

I'm confused.

Hmm, weird problem...I'm beginning to wonder whether that track is *supposed* to sound clipped, and it's only with your eq preset that the overall sound is clean enough to hear the harmonics that clipping creates!

The other distinct possibilty is that there is a bug in the player software that calculates the coeffiecients used in an eq section. If this is truly happening in the DSP and only when your cut-only eq is present, then this has to be the prime suspect. (Given that you've already eliminated my bass/treble hack as a cause - thanks) And remember that there was a bug fix in beta12 to fix the channel lock issue - if I was working for SB doing SQA on beta12, the first thing I'd do here would be to try to reproduce it on beta11...But there is the possibility that if it is a bug in the player, then it has existed forever - the SAA7705 programming manual does not appear to give us the format of the bytes for the eq sections, but refers us to EQU, a program used to calculate them. I'm guessing that this EQU program wasn't distributed as an ARM binary, and that the code that the player uses was specifically coded for the Empeg.

The track's got a lot of bass, but it's not supposed to distort like that.

I'm just thinking there's something weird in my settings that's making certain very narrow frequency bands boost instead of cut.

Ok, *Assuming* for a minute that there is an obscure bug in the coefficient calculation, it's quite possible that it's an integration bug eg:

The player thinks that one range of values is valid.
The DSP thinks that a different (lesser) range of values is valid.

Now for the most part, this kind of bug isn't apparent in everyday use, as most values fall in the DSP's range. But once you try to extend beyond that range then the value that you give the DSP isn't 'valid'. In the way that the eq sections are implemented in the DSP, I can tell you that one word (16 bits) appears to control gain, Q-factor and the 0/-6dB scaling. So it is possible that if there was a bug of this nature, then it would be possible that an invalid Q factor may very well affect the gain...

With all this sheer speculation in mind, take a look at band 4. Try reducing the Q factor to 5 and see what happens.

With all this sheer speculation in mind, take a look at band 4. Try reducing the Q factor to 5 and see what happens.

When the Q factor is below 1, there is no distortion. The distortion slowly increases as I push it past 1.4 or so.

This doesn't indicate a bug, since that's a bass frequency and I'm reducing other lower bass tones by decreasing the Q.

Although I will mess with the other settings more and see if there's a particular one that does it like an on/off switch. Good idea...

Interesting. It seems to be related to that first band somehow.

If I change that first band so that it's centered around 100-150 hz with a high Q (say, 20) and decrease it to below -27db, then the distortion goes away.

But if I leave that band as-is and make the SAME correction using one of the two unused bands (9 or 10), then it doesn't fix the distortion.

Not sure if that indicates a bug, because there's not a sudden moment when the distortion turns on or off. It's a gradual thing. And by messing with that frequency, I'm changing the response curve of the EQ overall...

At this point, one might be inclined to say that it's just the S/N of the unit, but with all bands flat, which is much much louder, there is no distortion at the same volume setting. Is there something I'm not understanding about how DSPs handle equalization? If I'm only cutting frequencies, then in theory, nothing should be getting louder, right, and therefore not distorting?

Hi,

How sure are you you're hearing clipping and not harmonic distortion? How sure are you the song doesn't just sound like this when highlighting certain frequencies as you've done? I think if you want a more conclusive test you have to try generate some reference tones with a reliable source and feed them into the player. The tones should be generated at different volume levels (-20, -15, etc). If it's fed in the form of an MP3, I'd suggest encoding them at 320 kbps just to be safe. Otherwise WAV if you have spare space for it.

And then to monitor the output, I wouldn't use headphones or an amp, unless you're 100% sure the distortion isn't occuring there. I think pulling it in via the line in on your PC is a good idea. Better yet, if you have or know somewhere that you can borrow a scope and other test equip...


Regards,
Aragon

The EQU program is a DOS prog Not sure how the empeg player software implements the equaliser though. They might have a big table of possible values or they might calculate it themselves on the fly.

According to the spec the equaliser when set to a flat response just has all of the gain of each section set to 0. This would mask any mistake or bug in the DSP or other coefficient values.

- Trevor

Greetings!

It seems to be band 1? Perhaps try swapping out that frequency into another band of the EQ. Yes, it shouldn't matter, but it will eliminate the first band as a source of problems as opposed to the frequency itself. I also note that you are centered around 50Hz. I tend to get a little more nervous around the joyous 60Hz area. (US AC power runs at 60Hz for those not in the US.) Could there be something in the original recording or in the audio stream that is "adjusting" the audio around 60Hz? If I remembered my math, I would try to figure out where the 1.4Q setting you notice the clipping fall off is, frequency wise...

Yeah, try to reproduce band 1 in band 9, and set band 1 to zero..That'd be a really obscure bug if that is indeed the issue.

One other thing I've noticed is that you have 2 bands with low Q factors and a high -ve gain. These bands will be knocking a huge chunk out of your sound...and maybe your listening volume with your eq active is pushing the DSP into clipping. But you haven't stated what your typical listening volume is, and I still wouldn't expect it to even contemplate clipping below 0dB, and if you did the flat eq test at that same volume then I'd expect it to still clip.

But you haven't stated what your typical listening volume is

In the car, I've adjusted my gains low enough so that I can crank it all the way up to 0db if the mood strikes me. Usually, I'm at -10db to -5db if I'm alone. However there are times when I do want to turn it all the way up to 0db.

How sure are you you're hearing clipping and not harmonic distortion?

Not sure.

How sure are you the song doesn't just sound like this when highlighting certain frequencies as you've done?

If you listened to the clip normally and then listened with the adjustments I've made, you can tell pretty obviously that it's not just a different sound, it's actually a distortion. It's not just a subtle difference in the bass, it sounds like a machine gun popping off.

Yeah, try to reproduce band 1 in band 9, and set band 1 to zero..That'd be a really obscure bug if that is indeed the issue.

Oh wow, now this is interesting.

When I reproduce the details of band 1 in band 9, then I set the gain parameter on band 1 to zero, the distortion gets significantly worse. But at worst, I should get the exact same sound out of either configuration, shouldn't I?

Something is fishy!

Agreed. You should get the exact audio as you had before. If it is worse, that implies that with every additional band added, it is a cumulative effect...

This is a little off the topic, but I had a problem with this kind of thing with my onkyo home stereo. When I would be in the dolbypro logic mode with the add on onkyo eq on I would get all kinds of distortion just like what your describing with the rio. When I would have the eq off it was fine or if I turned the dsp off and leave the eq on it was fine, the eq settings were all almost a flat line too with the mids and bass down a little. I tried a technics eq with the same settings and the problem went away. It was very strange when there would be a loud boom in a movie or music the bass would sound like it was cracking.

p.s. Tony... could you send a copy of beta 12 this way

Ah, equalizers. Man has tried to solve the mystery of equalizers since the dawn of time. The perplexing enigma of the equalizer is more daunting than the secret of life. The best one can do is pick a random setting or make a nifty little design with the sliders and hope for the best.

Man has tried to solve the mystery of equalizers since the dawn of time.

Reminds me of the Smith and Jones sketch: "For? What do you mean, what's it for? It's a graphic equaliser! It, er, it, um, er, it equalises yer graphics, dunnit?"

Peter

Heh, yeah.

Back on topic: Any clue as to what's going on?

Not really.

Are you able to reproduce it on other players?

Another thought: have you contacted Stu about running a test through his digital output board? Maybe we can narrow down where the problem is, or see it more clearly in digital.

I will try his clip with the settings through the digital out later today. Right now my Rio is stripped down to the bare case to permanently install the three digital out connectors.

Stu

I will try his clip with the settings through the digital out later today.

Thanks. Can you also try it with the analog out and see what you get?

My theory is that you will notice the problem on the analog outputs, but not on the digital output. This is just a theory though.

Remember that you have to hand-enter each of the settings in both the left and right EQ in "independent" mode since you don't have beta 12. Sorry it's such a pain.

Tony, you really should eliminate the possibility that the bug-fix in beta12 has somehow introduced this problem. Do you have a spare running beta11 that you can try?

BTW, I'm going to be incommunicado for the next 9 days, but if you haven't found the cause by then, I'll look into what can be done in the kernel to help debug this.

Tony, you really should eliminate the possibility that the bug-fix in beta12 has somehow introduced this problem.

Agreed. Since everyone else is running Beta 11, I was hoping that posting the settings could kill two birds with one stone by verifying the problem exists at all, as well as checking that it was present on beta11. If Maczrool gets a chance, this will take care of that requirement.

How do I enter "independent" mode with the EQ? I can't find it in the FAQ. I've got the player put back together and the "snippet" loaded. Let me know and I'll run the test.

Stu

I believe that "dependent" and "independent" correspond to "locked" and "unlocked" in the equalizer, respectively. But then, I'm only running 1.03, and don't know if it's different in 2b11.

I guess I need step-by-step instructions on how to get the independent EQ enabled. I never really use the EQ, because the Rio is not my main headunit. If someone knows where they are or could write some that would be helpful.

Stu

I guess I need step-by-step instructions on how to get the independent EQ enabled.

When in the equalizer, press the DOWN button on the front panel to open the menu. There should be a "Set Locks" option on this menu. That lets you choose "Independent" or "Locked L/R". This is in the 2.0x software of course.

I've done the test. The clipping is there on both analog and digital outs. If you want, I can send you a WAV file of the actual output from the digital outs. PM me and I'll send it out to you.

Hope this helps,

Stu

The clipping is there on both analog and digital outs.

Interesting!

And you have verified that it goes away with a flat EQ?

You can e-mail the WAV file to [email protected] at your convenience. I'm now broadband at home so the file attachment size is not a problem. Will this WAV file be some kind of a genuine digital grab from an S/PDIF input?

Thanks for doing this for me. I will run the WAV file through SpectraLAB and see what it tells me about the frequency distribution.

Yes, the clipping mysteriously vanishes with the Flat EQ setting, both on analog and digital outs. Yes, I will send the digital output from the coax S/PDIF to my digital sound card on my Mac and capture the output to hard disk and save as WAV, so it should be a "genuine digital grab."

One thing I must mention about the WAV file I will be sending you is that it may be 1dB quieter than the original MP3 due to something funny with the I2S outputs (the signals feeding my digital out board) on the main board of the Empeg. I have discovered that the volume must be reduced by 1 dB (from 0 dB) to remove some mild clipping during peaks in tracks. I know it's nothing our digital out circuitry is doing, because I've tried I2S (which bypasses everything but the line driver IC) and both optical and coaxial outputs with the same clipping at 0 dB. I've also tried different external DACs, with the same result.

Stu

If I recall correctly, the sample settings I gave you will produce the distortion even at -1db so that's fine.

Did you get a chance to play with the WAV file I sent you?

Stu

Unfortunately, oronet's mail and web servers went Tango Uniform yesterday, so I can't get my email at the moment. If your mail didn't bounce, it's probably sitting in my inbox on their server, and whenever it comes back up again I'll be able to retrieve it.

Okay, got the files onto my hard disk. Haven't listened to them yet. But in your accompanying email you mentioned that you noticed the distortion on only one channel.

Wonder if this has something to do with the Beta11 L/R bug? I think I'm hearing it in only one channel as well, but I'm running a beta that's not supposed to have the bug...

(And I did do all of the corrections in beta12, so in theory I should not have been saving a preset with beta11.)

Okay, yeah, that WAV file clearly shows the distortion on the right channel, encoded right in there.

Analysis of the file shows nothing special, other than that the file has a lot of bass. There's notable peaks at 32hz and 64hz, but nothing special about that. Correcting either of those frequencies downward sharply doesn't fix anything.

Wait a minute. Those are multiples of each other, aren't they? I wonder... Hang on...

WHOA, weird.

Okay, so I have this idea. The bass notes have peaks at 32 and 64 hz. So I think to myself, what if the problem is actually at 16hz? What if it's a harmonic of 32+64 going wonky?

Well, the EQ doesn't go down to 16. But it goes down to 18. So I take one of the two unused bands, set it to 18hz, give it a fairly high Q (20.0), and start correcting downward. At -5db, the music is still distorting, sounding like it always has. At -6db, things go NUTS, the sound distorts TERRIBLY on all frequencies. At -7db to -8db, there is no output except for the occasional bit of distortion/static on the heaviest downbeats. At -10db, there is no output at all from the player.

Wonder if this is related?

How do these peaks in my WAV file compare with the peaks in a WAV of the clip that has not been run through the Empeg's EQ?

Stu

Oh, and by the way, a direct spectral analysis of the player's outputs shows that I'm not being bitten by the L/R EQ bug as the two channels have an essentially the same spectral signature.

How do these peaks in my WAV file compare with the peaks in a WAV of the clip that has not been run through the Empeg's EQ?



Purple=Your supplied WAV file.
Green=Original, unmodified audio file without the Empeg being involved.

Funny how, apart from amplitude, they don't look that different, considering how severe some of the EQ corrections were.

I also don't know why yours has a greater amplitude, when all of the EQ corrections were cuts instead of increases.

INTERESTING...

I think it is some kind of a bug, either in the DSP or in the programming that goes to the DSP.

All right. Start with a flat EQ, any volume setting, any music....

Take the first band and set its frequency to 18hz (not khz) and its Q to 20.0.

Decrease the gain (with a fixed Q) and you will hear:

-6db= SEVERE Distortion in the RIGHT CHANNEL ONLY, while the left channel is fine.
-7db= Right channel gone, left channel fine.
-27db= Both channels fine again.

Of course, no one outside of the Alpha team need mess with this, as it's on internal beta 12 which is the version that's not supposed to have L+R bugs...

/me posts to internal bug list...

/me confused.

Tony, are you saying that this is a beta12 bug only?

I think it's in both beta12 and beta 11, and it's a different bug than the one-channel-only-gets-the-Q bug in beta11.

On a possibly related note. I was hoping to use a 18Hz high Q cut as a "poor mans" subsonic filter. Running pink noise through 2b11. comparing the input (reference) and output using various PC&soundcard 2 channel analizers (SpectroLAB, SmaartLive, & EFT). I noticed that below around 40Hz the EQ stoped working, meaning the output trace just floated back up to match the refence level (averaging on) as I selected a center frequency below that point. It was in the High 30's I think. Can't post results, didn't save em sorry. I say "around 40" only because I don't remember, but there was a consistant frequency below which this behavior occured, and was consistantly reproducable in both 2 and four channel, independant and locked modes. I was using only the first band of the parametric, all others were flat.

This led me to think the only way to get a subsonic High Pass filter would to use an fmod (they come in 20Hz and 50Hz, -12db/octave) into the amp. An unfortunante solution because my starting parameters (which I'd hoped to tweak) were more like -24dB @ 25 Hz.

This was my first pass of playing with the Rio "on the bench" (PC). Then am now embarked on my install (dash, subs, amp, cap, etc.). I plan to retry this again and in the car when thats complete. And will reproduce and post more detailed results.

Michael

Yes, I've used F-mods in the past (when my amps didn't have crossovers built-in, before I bought newer amps) and they worked quite well.

This is an interesting note about the lower limit for the EQ, and would explain why I was never able to make it satisfactorily work as a high pass filter. I wonder if this is a limitation of the DSP, or if it's a bug in the code?

"... after I did my RTA adjustments ..."

What are some of the cheapest ways to RTA a system? Renting an audio analyzer is mucho $$$ (~150 per day from a cursory search) which would limit me to one day. Any software based solutions that aren't very expensive would be nice.

Cory

I did it with a downloadable 30-day trial version of SpectraLAB and a Shure SM-58 microphone being run through an XLR-to-minijack convertor. The SM-58 was borrowed from a musician friend, so I essentially did it for free.

Oh, and DON'T expect one day to be enough. I've gone through several passes on my system, tweaking different things and learning a little more each time.

Thanks for the pointer. I've got an Electro-voice ND257 mic that hasn't seen any use for over a year... I fully expect to spend a couple of months fine tuning it. Funny how high powered and really high quality audio can expose a LOT of issues...

Anyone else interested in this approach should start out at http://www.soundtechnology.com/LAB432v17.EXE.

You might want to check and see if that EV 257 has a pre-defined response curve already built-in to the software. The reason I chose the SM58 was because there was already a template for it in SpectraLAB.

I actually called up my musician friend and read him the list of built-in mic templates, and he stopped me on the SM58 and said, "oh, I've got a bunch of those".

Reminds me-- I gotta return it to him one of these days.

hey tony, did that software to automatically change your EQ with a mic ever get released? i remember you talking about ti around the time of the bay area meet

The idea of writing such a piece of software was discussed, but non one ever wrote anything. Remember that the input sampling rate of the player's mic input would preclude the ability to adjust frequencies over about 11khz, so it would only be about a 50% solution, even after lots of work.

If the EV 257 is not included, I wonder if you can input a response curve on your own? EV ships many of their mics with individual response curves (a nice by-product of good QC). The more I think of it, entering one in manually might be more trouble than just borrowing another model from someone. I'll check tonight after I mow the lawn...

Yes, you can enter your own response curves in SpectraLAB. They give detailed instructions. I found it easier to just use an existing template instead of second-guessing myself wondering if I'd gotten it right.

And for more details about the procedure, make sure to read my original thread on the topic.

Someone else also posted about the Praxis RTA software, which doesn't have a 30-day limit. Not that I've used either one; just another option.