Topic: Edirol UA5 distortion *before* 0dB???

[poorlyconditioned]

You may have read my hackery with the UA5, changing opamps and such.

I decided to verify my results before taking it out into the field.  So, I used a "sweep", a sinewave that goes from 20 to 20K over about twenty seconds.  Well, I noticed some harmonic distortion (multiples of the base frequency) with anything above -6dB or so. 

Demonstration: To isolate the problem I focussed on one frequency.  I put a *single 1600Hz sine wave* onto a Creative NJB3 and put line out into *line in on the UA5*.  I used a laptop USB to record from the Edirol UA5 for approx ten seconds.  Then I did an FFT of the recorded wave using the Adobe audition program.

Here is the recorded wave when the input to the UA5 is at approx. -3dB.  Notice the secondary peaks at multiples of 1.6kHz.


Here is the same wave input at -6dB.  The peaks are mostly gone:


The problems seemed to go away below -12dB or so.  FWIW I noticed no such probems running a MD line input near 0dB.

OK, here are the files used:
Input (sinewave): www.cs.uwaterloo.ca/~mannr/UA5/tone1600.flac
Output of UA5 (-3dB): www.cs.uwaterloo.ca/~mannr/UA5/ua5-tone1600-3dB.flac
Output of UA5 (-6dB): www.cs.uwaterloo.ca/~mannr/UA5/ua5-tone1600-6dB.flac
Output of UA5 (-12dB): www.cs.uwaterloo.ca/~mannr/UA5/ua5-tone1600-12dB.flac
Output of UA5 (-18dB): www.cs.uwaterloo.ca/~mannr/UA5/ua5-tone1600-18dB.flac

Any guidance welcome!  Note that I used line input, so this should have nothing to do with my mods.  (And I tried it with and without mods too, because that is what I thought originally.)

  Richard

[keith]

Richard,

You refer to various dB levels, i.e. you say that the plot you show is with an input of "-3dB".  What is that with reference to, are you talking about dBu, dBV, or something else?  Since a decibel is a relative measurement it is always with reference to some level.

It definitely looks like the Edirol is beginning to clip - I wonder if the front end is designed with a different max level than your MD recorder?  What do the specs on the 2 devices claim for max input levels?

-Keith

[poorlyconditioned]

Richard,

You refer to various dB levels, i.e. you say that the plot you show is with an input of "-3dB".  What is that with reference to, are you talking about dBu, dBV, or something else?  Since a decibel is a relative measurement it is always with reference to some level.

It definitely looks like the Edirol is beginning to clip - I wonder if the front end is designed with a different max level than your MD recorder?  What do the specs on the 2 devices claim for max input levels?

-Keith

When I say 0dB and so on, it is with respect to the UA5 *digital output*.  So, to get -3dB, I just a moderate signal into line in (from a NJB3 line out BTW), and turn the trim pot on the back of the UA5 until the signal coming out is at -3dB.  That is all.  So, I'm not saying anything about actual analog levels here.

Note: I get problems with both mic and line in, so it must be something at, or near the ADC...

  Richard

[poorlyconditioned]

Richard,

You refer to various dB levels, i.e. you say that the plot you show is with an input of "-3dB".  What is that with reference to, are you talking about dBu, dBV, or something else?  Since a decibel is a relative measurement it is always with reference to some level.

It definitely looks like the Edirol is beginning to clip - I wonder if the front end is designed with a different max level than your MD recorder?  What do the specs on the 2 devices claim for max input levels?

-Keith

When I say 0dB and so on, it is with respect to the UA5 *digital output*.  So, to get -3dB, I just a moderate signal into line in (from a NJB3 line out BTW), and turn the trim pot on the back of the UA5 until the signal coming out is at -3dB.  That is all.  So, I'm not saying anything about actual analog levels here.

Note: I get problems with both mic and line in, so it must be something at, or near the ADC...

  Richard

OK, I'm going to reply to my own post here...

After talking with some people who know more than me, the conclusion is, that for cheap ADC at least, you shoud *not* run near 0dB.  Common sense would say get more signal and less noise, but it appears this ADC works best at say -6dB.  I'm not sure how to do this at shows though.  The CLIP light comes on at -3dB.  I guess you should get the clip light and then back off.

Just to reiterate, I observed secondary peaks at -40dB when the input signal was at full scale.  This is 1% distortion!  So, even if you run at -6 or even -12 and boost in post, you're going to do better than that.  Now that is distortion at the peaks only, but distortion none the less.

Now I don't know how other ADC's compare, but I do know that my MD has no such distortion at full scale!  The amps might not be as good, but the ADC seems alright.  I'm going to try my NJB3 line input as well.

I guess the moto is: trust your ears, but also run some tests!

  Richard

[Brian Skalinder]

Just to reiterate, I observed secondary peaks at -40dB when the input signal was at full scale.  This is 1% distortion!  So, even if you run at -6 or even -12 and boost in post, you're going to do better than that.  Now that is distortion at the peaks only, but distortion none the less.

I'm not very technical, so I'll ask:  might this be related to your associated hackery and not necessarily the stock design and ADC?

[poorlyconditioned]

Just to reiterate, I observed secondary peaks at -40dB when the input signal was at full scale.  This is 1% distortion!  So, even if you run at -6 or even -12 and boost in post, you're going to do better than that.  Now that is distortion at the peaks only, but distortion none the less.

I'm not very technical, so I'll ask:  might this be related to your associated hackery and not necessarily the stock design and ADC?

I tested with a stock unit too.  Same problem.  Same problem reported with "cheap" soundcards too.  Don't drive near 0dB is the message I guess.  I don't know whether pro gear (eg., Grace V3, etc) is subject to this or not.

  Richard

[SparkE!]

Still, this almost has to be a problem with the analog feed to the A/D.  The A/D itself should be very linear.  Otherwise they would not be able to claim the THD specs that they quote for a full scale signal.  I've been wondering about this myself.  I've noticed that the gain settings on the UA-5 tend to be pretty forgiving, like there is some soft clipping going on when you get close to the rails.  Soft clipping would give you the type of distortion you see in the 0 dB spectrum.

[poorlyconditioned]

Still, this almost has to be a problem with the analog feed to the A/D.  The A/D itself should be very linear.  Otherwise they would not be able to claim the THD specs that they quote for a full scale signal.  I've been wondering about this myself.  I've noticed that the gain settings on the UA-5 tend to be pretty forgiving, like there is some soft clipping going on when you get close to the rails.  Soft clipping would give you the type of distortion you see in the 0 dB spectrum.

I think it is "soft clipping" of some kind too.  I don't know why, bit it seems OK up to about -3dB.  So, my advice is to *never* let the clip light come on even though it looks like you're 3dB down in the recording.

It seems to me like some kind of bug in the design.  An ideal ADC would go right up to 0dB.  And in fact, I've got some docs on the chip (the "evaluation board" for the chip) that show a plot at -0.2dB!  But I think it is a bug perhaps in the Edirol implementation, either the setup of the ADC or the Cirrus resampling chip after it.

  Richard

[aberg]

Richard, that's great info to know, since I think most people were under the assumption that seeing occasional clip indications was the ideal setting.

[it-goes-to-eleven]

I'll try and do some tests of this with gear I have.  Of course I often use usb>ua5 for doing these tests.

Baudline is a Great free tool for doing noise analysis but it is linux only.

[SparkE!]

Richard, what type of op amps did you substitute into your box?

[poorlyconditioned]

I'll try and do some tests of this with gear I have.  Of course I often use usb>ua5 for doing these tests.

Baudline is a Great free tool for doing noise analysis but it is linux only.

Thanks, please keep us posted!

Can you give a brief description and URL to "baudline"?  Or is that "bodeline"?  Anyway, I do run Linux on my workstation computers.  I only use Windows for audio stuff.

  Richard

[it-goes-to-eleven]

Absolutely love it.

http://www.baudline.com/

[poorlyconditioned]

Richard, what type of op amps did you substitute into your box?

I finally decided on a BurrBrown OPA2228 for the first two (mic pres), and then OPA2134 for the next two.  I forget the numbers on the circuit board but the first two are right beside the XLR connectors.  My guess is that those are the most significant.  I chose OPA2228 because they are optimized for gain of >=5, which is good.

I'm kind of bummed out on the ADC though.  While the OPA's may reduce noise a bit, I really think the distortion is a much more significant issue.

  Richard

[SparkE!]

Richard, I think that the problem you are having is that neither of those op amps have rail-to-rail outputs.  The OP2134's output will swing closer to the negative supply than it will to the positive supply, so when you run too hot, you'll see more even harmonic distortion than odd harmonic distortion in the lower order harmonics.  I see evidence of that in your plots.  You may want to find an op amp that has rail-to-rail output capability and low noise.  I think that the op amps used in the stock UA-5 design also have this issue.

FYI, the 2134 will get within 2.2 V of the negative supply and to within 2.5 V of the positive supply when driving a 600 ohm load.  It will get to within .5 V of the negative supply and to within 1.2 V of the positive supply when driving a 10k ohm load.  I don't know what impedance is seen looking into the A/D circuit, but I suspect that it's probably closer to the 10k ohm scenario.

[poorlyconditioned]

Richard, I think that the problem you are having is that neither of those op amps have rail-to-rail outputs.  The OP2134's output will swing closer to the negative supply than it will to the positive supply, so when you run too hot, you'll see more even harmonic distortion than odd harmonic distortion in the lower order harmonics.  I see evidence of that in your plots.  You may want to find an op amp that has rail-to-rail output capability and low noise.  I think that the op amps used in the stock UA-5 design also have this issue.

FYI, the 2134 will get within 2.2 V of the negative supply and to within 2.5 V of the positive supply when driving a 600 ohm load.  It will get to within .5 V of the negative supply and to within 1.2 V of the positive supply when driving a 10k ohm load.  I don't know what impedance is seen looking into the A/D circuit, but I suspect that it's probably closer to the 10k ohm scenario.

Yeah, that might be true, but I saw the problem *with the original UA5* too.  I measured the voltage at the ADC pin at about 1V peak-to-peak, which is attainable by any opamp from at +/-5V supply.

I encourage others to take up this challenge if you know some electronics, but I honesty think something is broken in this circuit.  The solution *for now* is just to run -3 or even -6dB down and boost a bit in post.  I find it crazy in a way that we (tapirs) pay so much for gear and never check it out, but the companies don't either!  Just look at all the problems with the microtrack...

  Richard

[SparkE!]

Yeah, that might be true, but I saw the problem *with the original UA5* too.  I measured the voltage at the ADC pin at about 1V peak-to-peak, which is attainable by any opamp from at +/-5V supply.

Yes, and the NEC uPC4570 that the original UA-5 uses also does not swing rail-to-rail.  In fact it's 3 V off the rails with a +/-15V design.  It's going to be really soft with a +/-5V supply like is used in the UA-5.

Also, I checked on the input impedance of the AK4524 and it is typically 10k ohms, so you will get asymmetric soft clipping with the OPA2134.

I've long suspected that the main improvement in sound on a modified UA-5 comes from use of rail-to-rail op amps and the lower harmonic distortion that results. I think this will be much more noticeable than than the noise level improvement you get from going with a lower noise op amp.  For choosing op amps for this application I would search for op amps that have rail to rail outputs, high slew rate, high gain bandwith product and low noise, in that order of importance.

[Chuck]

AD8620 - A good choice for the A/D buffer.
AD712 - I've seen Doug Oade comment how much he like this for it's "warm" sound.

[aberg]

AD8620 - A good choice for the A/D buffer.
AD712 - I've seen Doug Oade comment how much he like this for it's "warm" sound.

That's cool, except for those of us with AKG's or Neumann's.....

[poorlyconditioned]

AD8620 - A good choice for the A/D buffer.
AD712 - I've seen Doug Oade comment how much he like this for it's "warm" sound.

Cool +++++++++++++++T

Now we're talking.  OK, I've got all those chips...

So, I'll put the buffer as AD820.

Before that there are three chips: one chip (a pair of amps) for the left, one chip for the right and one differential output (both L and R use the same chip).

So, would you put an AD712 right on the input, or as the differential, or as all of the above?

  Richard

I'm still working out this ADC though.  There must be something wrong, no???

[poorlyconditioned]

AD8620 - A good choice for the A/D buffer.
AD712 - I've seen Doug Oade comment how much he like this for it's "warm" sound.

That's cool, except for those of us with AKG's or Neumann's.....

Actually, from what I hear the Neumans want transparent and the AKG want warm.  So, I'll try the warm chips with my AKGs

  Richard

[Chuck]

I think the AKG C-480's sound good with a W-mod UA-5.
I don't have the background to be able to pick the right chips. I know the headphone amp builders really like the AD8620 in unbuffered amps and I've heard that it works well as an A/D buffer.

[SparkE!]

AD8620 - A good choice for the A/D buffer.
AD712 - I've seen Doug Oade comment how much he like this for it's "warm" sound.

Cool +++++++++++++++T

Now we're talking.  OK, I've got all those chips...

So, I'll put the buffer as AD820.

Before that there are three chips: one chip (a pair of amps) for the left, one chip for the right and one differential output (both L and R use the same chip).

So, would you put an AD712 right on the input, or as the differential, or as all of the above?

  Richard

I'm still working out this ADC though.  There must be something wrong, no???

You'd use the AD712 as the input amps and the AD8620 everywhere else.  That's if you want the warm front end.  If you want a more transparent front end with lower noise, use the AD8620 everywhere.  The AD8620 was designed for use with +/-5V power supplies and it clips symmetrically at just 1V off the power supply rails.  If you make the mods, I'm betting that your distortion problem will go away.

[poorlyconditioned]

AD8620 - A good choice for the A/D buffer.
AD712 - I've seen Doug Oade comment how much he like this for it's "warm" sound.

Cool +++++++++++++++T

Now we're talking.  OK, I've got all those chips...

So, I'll put the buffer as AD820.

Before that there are three chips: one chip (a pair of amps) for the left, one chip for the right and one differential output (both L and R use the same chip).

So, would you put an AD712 right on the input, or as the differential, or as all of the above?

  Richard

I'm still working out this ADC though.  There must be something wrong, no???

You'd use the AD712 as the input amps and the AD8620 everywhere else.  That's if you want the warm front end.  If you want a more transparent front end with lower noise, use the AD8620 everywhere.  The AD8620 was designed for use with +/-5V power supplies and it clips symmetrically at just 1V off the power supply rails.  If you make the mods, I'm betting that your distortion problem will go away.

OK, the distortion problem did *not* go away.  I put 8620 as the buffer and as the differential amp summer.  I put AD712's on each mic input.  Distortion is still there.  I challenge anyone who has Oade mods to check it out, play my 1600Hz file from line out into your line in (or mic in with gain all the way down).  I think everyone is going to experience this distortion.

Now, I noticed if I go down to -4.5dB, the biggest secondary (3200Hz) goes from about -40 down to -80dB, so it is pretty far down.  At that level, it could be due to distortion from the NJB3 for all I know.  But do not go over -3dB on the input is my suggestion.

OK, I'm curious.  Does anyone know how the clip light works?  It comes on at 3dB. Maybe somehow the chip *is* really clipping there?  Also, I saw the AKM docs for their development board, and they show a spectrum, with great performance, at -0.2dB.  So it should work right at the top.  So my guess is there is a problem in Edirol's implementation, either in setup of the AKM chip (gain settings maybe?) or maybe the sample rate converter after it.  Who knows at this point.

  Richard

[Chuck]

AD8620 - A good choice for the A/D buffer.
AD712 - I've seen Doug Oade comment how much he like this for it's "warm" sound.

Cool +++++++++++++++T

Now we're talking.  OK, I've got all those chips...

So, I'll put the buffer as AD820.

Before that there are three chips: one chip (a pair of amps) for the left, one chip for the right and one differential output (both L and R use the same chip).

So, would you put an AD712 right on the input, or as the differential, or as all of the above?

  Richard

I'm still working out this ADC though.  There must be something wrong, no???

You'd use the AD712 as the input amps and the AD8620 everywhere else.  That's if you want the warm front end.  If you want a more transparent front end with lower noise, use the AD8620 everywhere.  The AD8620 was designed for use with +/-5V power supplies and it clips symmetrically at just 1V off the power supply rails.  If you make the mods, I'm betting that your distortion problem will go away.

OK, the distortion problem did *not* go away.  I put 8620 as the buffer and as the differential amp summer.  I put AD712's on each mic input.  Distortion is still there.  I challenge anyone who has Oade mods to check it out, play my 1600Hz file from line out into your line in (or mic in with gain all the way down).  I think everyone is going to experience this distortion.

Now, I noticed if I go down to -4.5dB, the biggest secondary (3200Hz) goes from about -40 down to -80dB, so it is pretty far down.  At that level, it could be due to distortion from the NJB3 for all I know.  But do not go over -3dB on the input is my suggestion.

OK, I'm curious.  Does anyone know how the clip light works?  It comes on at 3dB. Maybe somehow the chip *is* really clipping there?  Also, I saw the AKM docs for their development board, and they show a spectrum, with great performance, at -0.2dB.  So it should work right at the top.  So my guess is there is a problem in Edirol's implementation, either in setup of the AKM chip (gain settings maybe?) or maybe the sample rate converter after it.  Who knows at this point.

  Richard

Hmmm... Interesting...

[it-goes-to-eleven]

I did some tests last night..  I'll try and post some fft images, etc.

I ran jb3 out >

 v3 AES > 722
 722
 UA5 USB > PC

Suspecting the jb3, I then re-did the test using a Squeezebox3 as the source.  I'd like to test the minime in there and the microtrack too. I would especially like to get data on a Mytek A/D (double good with a Mytek DA as the source).  It would be fun to test a few high end D/A's with the same A/D destination.

Most tests were done at 24/44. UA5 @ 16 bits.  For a couple cases I also did 16/44 on the 722.

In one case I ran the limiter on the 722 and kept increasing the levels until they were absurdly high. The integrity of the sine wave was good, even at extreme levels.  But when 'in' the limiter (above -6 dB), the harmonics increased.

Some portion of the harmonic is probably from the D/A source.

The harmonics in the V3 A/D source were lower than the 722.  But most who own both agree that V3 line > 722 sounds better than v3 AES > 722.  Go figure!  When recording shows, I record V3 line > 722 and also straight into the 722. I run my 722 levels hot. Would those soaring high female folk vocals that touch 0 dB sound better if they were peaked at -5 and raised in post?

Of course.. I now have too much data to make the summary a quick and painless task 

[poorlyconditioned]

I did some tests last night..  I'll try and post some fft images, etc.

I ran jb3 out >

 v3 AES > 722
 722
 UA5 USB > PC

Suspecting the jb3, I then re-did the test using a Squeezebox3 as the source.  I'd like to test the minime in there and the microtrack too. I would especially like to get data on a Mytek A/D (double good with a Mytek DA as the source).  It would be fun to test a few high end D/A's with the same A/D destination.

Most tests were done at 24/44. UA5 @ 16 bits.  For a couple cases I also did 16/44 on the 722.

In one case I ran the limiter on the 722 and kept increasing the levels until they were absurdly high. The integrity of the sine wave was good, even at extreme levels.  But when 'in' the limiter (above -6 dB), the harmonics increased.

Some portion of the harmonic is probably from the D/A source.

The harmonics in the V3 A/D source were lower than the 722.  But most who own both agree that V3 line > 722 sounds better than v3 AES > 722.  Go figure!  When recording shows, I record V3 line > 722 and also straight into the 722. I run my 722 levels hot. Would those soaring high female folk vocals that touch 0 dB sound better if they were peaked at -5 and raised in post?

Of course.. I now have too much data to make the summary a quick and painless task 

Excellent!  Looking forward to the results...

  Richard

[it-goes-to-eleven]

Richard's -6dB result on the left again..

First image is the 722 fed by the sb3 in line mode, 24/44.
Second image is the same with the 722 in low gain mic mode (a higher gain range that will add up to 55dB).
Third image is the V3 in 24/44 as fed by sb3 with output via AES to the 722.

[poorlyconditioned]

My UA5, as fed the signal by the jb3.  My efforts to get the levels exact were clumsy at best. So I tried to increment it up gradually and then pick out the section of interest from the sample.

I seem to have missed most of one channel. Not sure why.  Maybe bad cable seating.

Richard's result is on the left.  He was aiming for -6 dB. Baudline thinks it is higher than that.

Next image is basically the same but the source is the squeezebox3.

Hey, freelunch.  Those are great.  What I'm wondering is can you crank the input up really high, eg., -1 or -2dB?  I want to see how they all compare, namely the UA5, 722, and V3.  At -6dB it looks like V3 is better than 722 is better than UA5, but neither is terrible.  I mean, even -80dB down for the 3200Hz harmonic is not terrible on the UA5.  I want to see when the levels get way up there...

Thanks again,
  Richard