MT-32 vs. CM-32 et al.

[NewRisingSUn]

I only have the schematics for the first-gen MT-32 and the CM-32L (CM-64 LA Board).

CM-32L connects LA32's SD15-SD0 to DAC's BIT1-BIT16, as it should.

First-gen MT-32 connects LA32:D15 to DAC:BIT1 (sign-carrying bit), drops LA32:D14 (highest monotonous bit), then connects LA32:D13 to DAC:BIT2, LA32:D12 to DAC:BIT3, ..., LA32:D1 to DAC:BIT14, LA32:D0 to DAC:BIT15, and nothing to DAC:BIT16. Just as I described in the wikipedia article.

What makes you think SD0 is not used for data output?

[Cloudschatze]

What makes you think SD0 is not used for data output?

The service notes for the second-generation MT-32, LAPC-I and CM-64 all state that "the LA chip places the sound data on the data bus (SD1 - SD15) connecting to both the Reverb chip and DAC. (The sound data won't be output on the data bus (SD0) because the data is 15-bits data.)"

The first-generation MT-32 notes don't specifically mention whether or not anything is going out on SD0, however, they do state that the LA32 chip is generating 15 bits of data. If nothing is being sent on SD0, then the DAC is only receiving 14 bits of data, right?. Is there a way to see if anything is going out on SD0?

[NewRisingSUn]

Maybe they mean 15 bits without the sign bit. Does the D-50/D-550 use the same LA32 chip as the MT-32? Because that one supposedly has a resolution of 20 bits, and if it's the same LA32 chip, I couldn't imagine where the 15 bit limitation would come from.

Is there a way to see if anything is going out on SD0?

Simply hold the tip of an RCA plug against the correct DAC pin. Turn down the volume first, as +5V are pretty loud.

[Cloudschatze]

Maybe they mean 15 bits without the sign bit. Does the D-50/D-550 use the same LA32 chip as the MT-32? Because that one supposedly has a resolution of 20 bits, and if it's the same LA32 chip, I couldn't imagine where the 15 bit limitation would come from.

Is there a way to see if anything is going out on SD0?

Simply hold the tip of an RCA plug against the correct DAC pin. Turn down the volume first, as +5V are pretty loud.

They use a similar LA32 chip - the MB87316 in the D-50, versus an MB87316A in the MT-32. I have no idea what the internal differences are, but the I/O is the same for both.

Here is a link to the D-50 service manual:
http://dj.epo.free.fr/documents/d50/D-50_servicemanual.pdf

Check out what its doing with the lower 4 sound-data bits. Someone might have interpreted this as 20-bit resolution.


I'll give the RCA cable test a try, unless you've already done it.

[NewRisingSUn]

They use a similar LA32 chip - the MB87316 in the D-50, versus an MB87316A in the MT-32. I have no idea what the internal differences are, but the I/O is the same for both.

I wonder what would happen if one replaced the MT-32's "A" chip with the D-50's "A"-less chip...

Check out what its doing with the lower 4 sound-data bits.

I have absolutely no idea what it's doing there.

I'll give the RCA cable test a try, unless you've already done it.

No, my electric screwdriver doesn't seem to work anymore.

[Cloudschatze]

They use a similar LA32 chip - the MB87316 in the D-50, versus an MB87316A in the MT-32. I have no idea what the internal differences are, but the I/O is the same for both.

I wonder what would happen if one replaced the MT-32's "A" chip with the D-50's "A"-less chip...

I don't think I want to be the one to find out.

Check out what its doing with the lower 4 sound-data bits.

I have absolutely no idea what it's doing there.

External feedback-resistor circuit, perhaps?

I'll give the RCA cable test a try, unless you've already done it.

No, my electric screwdriver doesn't seem to work anymore.

[Cloudschatze]

Here are the results from the DAC testing adventure:

00 MT-32, SD0/DAC Bit 15 = "static"
01 MT-32, SD0/DAC Bit 16 = "static"

So, even though the 01 MT-32 service notes state that the LA32 chip isn't putting anything on SD0, this would seem to suggest otherwise. I thought that it might be from the reverb chip, but I still get a "signal" with dry tones.

Hmph. I really thought that I was going somewhere with this. Now I'm not sure what to think.

[BlueMax]

You guys' technical knowledge makes me feel unworthy of "geek" status. 

[NewRisingSUn]

00 MT-32, SD0/DAC Bit 15 = "static"

Yes, I measured the same on my already-opened first-gen MT-32. Meanwhile, I can't seem to find the highest bit on the PCB at all --- I know it isn't connected to the DAC according to the service notes, but I thought I might pick it up somewhere on the PCB component side...

Hmph. I really thought that I was going somewhere with this. Now I'm not sure what to think.

Maybe you could ask your contact at Roland Japan.

By the way, I made some recordings in "test mode". Those sine waves are hardly ideal ones! Consider:
Perfect 100,200,440 Hz sine waves
As generated by first-gen MT-32 test mode
Same, but with equalizing applied to highlight to imperfections

Now, do you think this is because the LA32 chip itself doesn't generate proper sine waves, or is it because of bad D/A conversion, or maybe even because of the deglitcher?

[Ari]

You guys' technical knowledge makes me feel unworthy of "geek" status. 

You and me both, man... 

 

[Tom]

 

Amen.

[jbltecnicspro]

You guys' technical knowledge makes me feel unworthy of "geek" status. 

You and me both, man... 

 

Haha - I'm just watching them sort all this stuff out from the sidelines, hoping that we'll get some sort of answer to... whatever it is that they're trying to determine...What is the original question, anyway?     Are you guys simply comparing the MT-32 to the CM-32L, or is there something deeper that you want to discover?

[Cloudschatze]

Meanwhile, I can't seem to find the highest bit on the PCB at all --- I know it isn't connected to the DAC according to the service notes, but I thought I might pick it up somewhere on the PCB component side...

SD14, right? You should see the following set of holes leading from the left of the LA32 PGA:

 o o o o o o
o o o o o x

SD14 is the "x". From there, it only goes to the reverb chip, whereas the other holes/bits can be traced to both the reverb chip and DAC.

Maybe you could ask your contact at Roland Japan.

No good. I think I frustrated them with my last question, to which they replied, "...basically, we support our domestic customers." I may bother Adrian Bruce about it. I'm not sure if he was directly involved in the MT-32/LA32 design, but then again, there is that "Tks to Masa + Adrian" ROM message...

Now, do you think this is because the LA32 chip itself doesn't generate proper sine waves, or is it because of bad D/A conversion, or maybe even because of the deglitcher?

I'm not sure. I'll record the same output from an 01 MT-32 later for comparison.


I am considering removing the PCM54HP DACs from my two 01 MT-32s. I'd like to install sockets and then drop in the higher-spec. PCM54KP DACs (possibly adding the trim circuit as well). There are a couple of DIY-audio forums that seem to suggest that the upgrade might be worthwhile. The chips themselves are prohibitively expensive though, at ~$32USD each, with a quantity 12 minimum order...

[NewRisingSUn]

Are you guys simply comparing the MT-32 to the CM-32L, or is there something deeper that you want to discover?

The missing DAC bit holds the meaning of life.

SD14 is the "x". From there, it only goes to the reverb chip, whereas the other holes/bits can be traced to both the reverb chip and DAC.

That's what I thought, too. But I'm not sure because if that is the bit whose lack of being connected to the DAC causes the "crackling" sound at high volumes, shouldn't I only hear a signal there exactly when a crackling sound is observed in the main output, because otherwise the bit isn't used?

I would assume that the four holes right to the "51" are the lowest four bits, then.

The chips themselves are prohibitively expensive though, at ~$32USD each, with a quantity 12 minimum order...

Hm. I could be persuaded to buy at most four. On the other hand, if we're talking about DIY modification, why not go all the way and use a completely different DAC (or even digital output)?

[Laust]

I am considering removing the PCM54HP DACs from my two 01 MT-32s. I'd like to install sockets and then drop in the higher-spec. PCM54KP DACs (possibly adding the trim circuit as well). There are a couple of DIY-audio forums that seem to suggest that the upgrade might be worthwhile. The chips themselves are prohibitively expensive though, at ~$32USD each, with a quantity 12 minimum order...

I looked into getting some PCM55s for comparison at some point, and balked at the price/quantity as well. Sadly, TI doesn't offer free samples for this device, only "samples to purchase".

It's not really surprising the output on the MT-32 isn't perfect (after all, it drops bit 14 and fudges with the rest), and I'm inclined to answer "all of the above"? to NewRisingSun's question   It would be interesting to compare with a D-50, and its test mode (including the DC-offset test/calibration which presumably provides a _proper_ sine wave as per the datasheet).

[Cloudschatze]

Here are the 01 MT-32 sine waves Sorry about the file size - my little recorder is fixed at 44.1kHz, and downsampling afterward would have produced poor results.

SD14 is the "x". From there, it only goes to the reverb chip, whereas the other holes/bits can be traced to both the reverb chip and DAC.

That's what I thought, too. But I'm not sure because if that is the bit whose lack of being connected to the DAC causes the "crackling" sound at high volumes, shouldn't I only hear a signal there exactly when a crackling sound is observed in the main output, because otherwise the bit isn't used?

It's definitely SD14. I've been thinking about this all evening, and have more questions than guesses at this point. I need to muse on it a bit longer.

Let me ask this in the meantime:

If the DAC in an 01 MT-32/CM was properly calibrated (no DLE, non-monotonicity, etc.), would it still crackle/clip? If yes, and assuming the service notes are correct, is the reverb to blame, given its 16-bit data, as opposed to the LA32's 15-bit data?

I would assume that the four holes right to the "51" are the lowest four bits, then.

0 - 3, right to left.

The chips themselves are prohibitively expensive though, at ~$32USD each, with a quantity 12 minimum order...

Hm. I could be persuaded to buy at most four. On the other hand, if we're talking about DIY modification, why not go all the way and use a completely different DAC (or even digital output)?

Relative ease versus the great unknown? 

The 01 MT-32 and CM-32/64 even have the trim circuit in place - all that's needed are three resistors and a potentiometer.

[NewRisingSUn]

Here are the 01 MT-32 sine waves Sorry about the file size - my little recorder is fixed at 44.1kHz, and downsampling afterward would have produced poor results.

Sounds about the same. Which makes sense, as the sound quality difference between 00 and 01 mainly concerns zero-crossing noise, which manifests itself mostly at low amplitudes. What we're hearing here must be something else.

It's definitely SD14. I've been thinking about this all evening, and have more questions than guesses at this point. I need to muse on it a bit longer.

I found the answer myself. Silly me! Of course, the bit will constantly change value even at low volumes because when the waveform crosses zero, all bits change their sign. For example, the tiniest volume step, from 0 to -1 (max would be from +32767 to -32768) results in hex 0000 to FFFF, or binary, all zeros to all ones. Indeed, SD14 sounds exactly like SD15 (the sign-carrying bit) except when a crackling sound occurs in the main output.

If the DAC in an 01 MT-32/CM was properly calibrated (no DLE, non-monotonicity, etc.), would it still crackle/clip?

Yes, because the crackling occurs within the LA chip itself. The RealWorld modification notes state that the crackling occurs on the 00 MT-32 if SD14 is used in the Partial sum, or if one particular Partial becomes too loud, because the LA32 chip only checks the Partial sum for overflows and soft-clips if necessary; it does not check the individual Partials for overflows.

The first variant of course doesn't happen on the 01 (or does it? See below.), but the second still can happen, and in fact it seems to happen more often, because the firmware seems to direct the LA32 chip to produce higher volumes (to get the same "double the digital volume" effect as the bit-shift in the 00 did, but without losing resolution), as I have written in the Wikipedia article. This also means that the crackling on the 01/CM is irrelevant of the MIDI main volume setting (set via sysex), because that one is applied at the Partial summing stage, when the overflow has already occured.

Check out this MIDI file from "Castles". It resets, then sets the MIDI main volume to the low value of just 20%, then plays a tune. Pay particular attention to the very last note --- even at this low MIDI main volume, it still crackles on the 01, while playing perfectly fine on the 00. The only way to fix this is to lower the volume of controllers 7 or 11!
(The crackling will typically occur at different places on subsequent playbacks because it largely depends on the current phase/polarity of the SQU/SAW waveforms, which depends on the arrival time of the MIDI note at the module relative to the LA32 chip's main clock.)

On the other hand, playing it at 100% main volume makes it crackle even more on the 01, so maybe there's something else going on. You know what? Maybe the 01 has the same bit shift like the 00, but implements it within the LA32 chip, which would mean that it outputs 15 bits after all, thus going along with the service notes. Why would they do this? Maybe to prevent people from modding the unit for better sound quality! Or am I already in tinfoil hat territory on this one?

If yes, and assuming the service notes are correct, is the reverb to blame, given its 16-bit data, as opposed to the LA32's 15-bit data?

No, why? The reverb return goes into a different time slot than the reverb send, and they're summed in analog.

The 01 MT-32 and CM-32/64 even have the trim circuit in place - all that's needed are three resistors and a potentiometer.

Tell me how much of a difference it makes. My assumption is, very little, but I'd be happy to be proven wrong on this.

[Cloudschatze]

Check out this MIDI file from "Castles".

I appreciate you posting this. I have a better idea of what is going on now, though the why still escapes me (with regard to why the design allows for such anomalies in the first place).

On the other hand, playing it at 100% main volume makes it crackle even more on the 01, so maybe there's something else going on.

How much might the analog output stages affect things? I don't want to suggest that it might be a problem (since I really have no idea), but the 01 MT-32 and CMs include a VCA after the low-pass filter, whereas the 00 MT-32 does not. (Nor does the D-10, for that matter...)

Edit: I wonder what the CPU is using to determine the CV value, and whether the VCA output at 100% is capable of overloading the op-amps that follow...

The 01 MT-32 and CM-32/64 even have the trim circuit in place - all that's needed are three resistors and a potentiometer.

Tell me how much of a difference it makes. My assumption is, very little, but I'd be happy to be proven wrong on this.

I gave Jim Williams a call. (If you search Google Groups for "PCM54KP," he's referenced as being responsible for several such modifications.) He told me that he stopped performing the upgrade because it didn't result in any noticeable difference in the output, despite the paper-specs. Instead, he suggests that the trim circuit will provide the greatest benefit (when properly adjusted with multi-thousand dollar equipment).

I told him that I planned on adjusting by ear. I guess we'll see how well that goes.

I don't plan on pursuing the DAC upgrade any further, but I am going to gather the components for the trim circuit. I'll order extras, in case you guys want to play along.

[Cloudschatze]

So, I'm reading about analog output circuits, and happen upon the following sentence:

"And, if you've ever experienced spurious AM-band signal detection by a bipolar input stage op amp, you'll be able to relate to all of this, for sure!"

EEs are so funny.



I received an e-mail back from Adrian Bruce. Apparently, Adrian Scott is the one referenced in the ROM; he being the sound designer, and Masa the product engineer. Contacting Masa might be worthwhile...

[Cloudschatze]

I'm ashamed to admit just how long it took me to come up with this, but here's what I've deduced regarding the VCA. (Perhaps you guys already knew this...)

In an 00 MT-32, the analog output is always at 100%. As a result, the continuous background hiss/whistling is always audible, regardless of the volume settings.

In an 01 MT-32/CM, the analog output is variable due to the VCA (which is controlled by the master volume parameter). As such, the background hiss is likewise variable, and at a master volume setting of zero, you get dead silence.

I'm not sure if it matters or not, but there you go.


Edit: Oh, and the D-10 substitues an analog slider for the VCA in the output circuit. (If anyone cared.)