MT-32 Advanced Info.

[Cloudschatze]

Written by Martin Sant, of Blue Ridge Music, many moons ago:

Ah, the MT-32; a great little box - for the money.  I dearly love
mine, even though it is kind of noisy.  I have gotten a lot of requests
for info on the MT, so I will attempt to share what I know about it.
Most of this info also applies to the D110, except the D110 has more
reverb modes, more PCM samples, battery backup, a card slot, and
killer drum samples.

First, the obvious.  The MT is a multi-timbral sound module that uses
Roland's now famous (infamous ?) sound generation method called LA
(Linear Arithmetic) synthesis.  This is really just a fancy name for
mixing PCM samples with 'normal' sawtooth/square wave kinds of sounds
While the timbres produced by the LA method sound complex, programming
the sounds is actually rather easy.

Of course, to change any of the MT's parameters, you need a computer
and the proper software; there are only a few things you can change
from the front panel.

The MT is logically divided into several sections.  First are the
synth parts.  There are 8 of these, each of which has its own midi
channel number, volume and pan setting.  Kind of like 8 synths and a
mixer in one box.

The next section is the drum section.  This area allows you to assign a
canned rhythm sound or any sound loaded into RAM to a particular note
number.  You can also set the pan, volume and turn the reverb on or off
for each of these sounds.

Now, all of the synth parts and the drum section draw from a 'pool' of
sounds arranged in banks of 64.  There are two banks of 'musical' sounds
in ROM (A and B), one bank of 'drum' sounds in ROM (actually only 30 of
these, abbreviated R) and one bank of RAM sound locations (called M for
memory).

You can get to the ROM A and B sounds in two ways.  The first way is to
select a part button and spin the knob.  The little micro inside the MT
dutifully loads the sounds into the part as the knob goes round.  You can
also use MIDI program change commands to pull up the sounds.  There is
a difference tho, 'cause when you use the program change commands to pull
up a sound, the MT uses an area called the patch map to get to them.

The patch map is an area of 128 locations, each of which corresponds to one
of the 128 possible MIDI program change numbers.  Each entry in the map
specifies a bank (A,B,R or M) and a number (1-64).  When you power up your
MT, this map is defaulted so that program change numbers 1-64 will 'point'
to ROM bank A and numbers 65-128 point to bank B.  Using a computer and
editor, you can change these map entries to point to any sound in any bank.
And, not only can you specify the bank and sound number, each map
entry allows you to specify the bend range, the keyshift (in semitones),
the polymode (whether the next note on cuts off the previous note or
triggers a new one) and the whether the reverb is on or off.  Note that
there is only one patch map.  If you send a program change to one part
on a certain channel and then send the same program change the another
part on another channel, they will both be set according to that particular
patch map entry.

The last group is the system section which controls the global things
such as master volume, tuning and the reverb.  The MT has four kinds of
reverb: room, hall, plate and discrete echo.  You can set the reverb time
from 1-8 and the level from 0 to 7.  Note that the reverb on the MT is
global.  You can turn the reverb on or off for each part or drum sound,
but you can't have one part set to one thing and another part set to
something else.

So all in all the MT is really very flexible in terms of setting things
up and pulling up sounds via midi.  Now I will try to break down the
parameters that make up a sound.

The smallest unit of sound generation is called a partial.  Four partials
are grouped together to form a timbre.  You can make a timbre using all
four partials if you want, or you can use just one or two.  There is a
section of the timbre refered to as the common parameters. This area
contains the 'switches' to turn on or off any of the four partials and
the partial structures (see below)  All together the MT can generate
32 partials at any one instant.  These are all dynamically allocated
between the synth and drum parts.

Let's look at a partial.  Each one of these is made up of
a Waveform Generator (WG), Time Variant Filter (TVF) and a Time Variant
Amplifier (TVA).  Each of these blocks has several parameters and a five
stage envelope generator associated with it.  The WG has controls which
affect the coarse and fine pitch, select the waveform (sawtooth, square
or PCM #) and control the amount of envelope effect.  The TVF parameters
include cutoff frequency, resonance, velocity sensitivity, several keyfollow
parameters and a five stage envelope generator (the last stage of the
envelope is fixed as the sustain level and release time).  The TVA has an
overall level adjustment, velocity sensitivity controls, a five stage
envelope and two programmable keyfollow controls.

Partials can be one of two types: Synth or PCM partials.  In the Synth
partials, all of the above mentioned blocks are active.  You can set up
a sawtooth on the WG, filter it with the TVF and adjust the volume with
the TVA.  In a PCM partial, instead of using the standard square or
sawtooth kind of waveform, a sampled waveform is selected.  Some of these
samples are looped so they sustain, while others only sound during the
initial attack of a note.  Also, in a PCM partial, the TVF has no effect.

As mentioned above, partials are combined into blocks called timbres.
Each timbre is made up of two pairs of partials.  These pairs are combined
by setting a parameter called a structure.  The structure tells the partials
what kind they are  (Synth or PCM) and how their output is combined.
Depending on the structure setting, the partials can be set up as 2 synth
partials, a PCM partial and a synth partial, 2 PCM partials, stereo PCM/synth
partials (one partial out the left and one out the right) and ring modulated
synth and/or PCM partials. (13 all total) Very flexible.  I should note here
that the MT firmware has a bug which does not allow the volume to be
controlled properly if structures 10-13 are used. (Maybe this is just a H/W
limitation) Cloudschatze: This bug was addressed in the final MT-32 ROM revision, 2.07.[/i]

Now for the MT insides:

The MT-32 uses an Intel 8097 microcontroller as the main CPU.  It has
a total of 64K of ROM and 32K of RAM.  The RAM is not battery backed, so
you have to reload all of your sounds and the patch map when you power up.
The CPU (and the operating system in the ROMs) controls the custom LA32
chip.  The LA chip has connections for the PCM sample ROMs (two 256Kx8 Toshiba
Mask ROMs on board rev 0 or one Hitachi 512Kx8 ROM on board rev 1) and the
digital reverb chip.  Contrary to popular opinion, the LA chip sends all it's
data to the reverb chip in digital format, not in analog form.  The
data is routed to the D/A converter (a BurrBrown PCM54 chip) and out
to 6 sample/hold chips.  With a few exceptions, this is the same circuitry
as in the D110.  (And probably the D10 and 20, but I don't have them so I
can't look inside)  For some reason, one of the data
lines from the LA chip to the D/A is LEFT OFF, killing the S/N ratio.  Also,
(as near as I can tell with the trusty 1240 logic analyzer) the sample data
in the ROMs has no data for this bit.  Note that the D110 runs ALL 16 bits
to the D/A which explains its better S/N ratio.

[NewRisingSUn]

For some reason, one of the data lines from the LA chip to the D/A is LEFT OFF, killing the S/N ratio.

Is this also the case on the CM-32L? If not, maybe that's the reason why the CM sounds cleaner than the MT-32, even though all the chips are reportedly the same.

[Cloudschatze]

For some reason, one of the data lines from the LA chip to the D/A is LEFT OFF, killing the S/N ratio.

Is this also the case on the CM-32L? If not, maybe that's the reason why the CM sounds cleaner than the MT-32, even though all the chips are reportedly the same.

I'll have to check on this later. Laust may know

The author of the above text also posted an MT-32 Noise Modification, to rid your MT-32 of "grunge".

Being from the Seattle area, mine will be grungy forever.

http://groups-beta.google.com/group/rec.music.synth/browse_frm/thread/846dcea068c4f670/ebcd3f5d0331b6cc

[Laust]

snip

Blah, and here I was hoping for something new and interesting, not a text that's been on the net for years...

[Cloudschatze]

For some reason, one of the data lines from the LA chip to the D/A is LEFT OFF, killing the S/N ratio.

Is this also the case on the CM-32L? If not, maybe that's the reason why the CM sounds cleaner than the MT-32, even though all the chips are reportedly the same.

Well, the CM-64 schematics show all 16-bits to the DAC, as do the schematics for the revised, MT-32  01 mainboard. The "left off" bit must have applied solely to the 00 mainboards.

I'm now wondering how much cleaner this 01 board sounds, and thinking that the noise level is probably very close to, if not exactly the same as that of the CM-series.

[Great Hierophant]

There is a guy on eBay claiming that the old-revision MT-32s without the headphone jack are superior because they use a "faster processor."  If so, this could be another reason why some MT-32 songs don't playback correctly on later modules if they also use a slower processor.  

Or maybe only the later MT-32s suffer from this problem and that is why some games experience sound buffer overruns with the MT-32 but not the other modules, i.e. the slower processor cannot process the midi data fast enough for the game.  Roland had other ways to cut corners from the later  CM modules.

[Laust]

Well, the CM-64 schematics show all 16-bits to the DAC, as do the schematics for the revised, MT-32  01 mainboard. The "left off" bit must have applied solely to the 00 mainboards.

I can confirm that bits 13-16 (didn't bother checking the lower ones) are indeed routed from LA chip and reverb chip to the DAC. I believe the samples only have 15 bits worth of resolution, but it should still improve the quality a bit (no pun intended), as all the processing the LA32 or the reverb chip does on those samples will be output in the full 16 bit resolution.

BTW, the schematics for the revision 0 board are quite confusing when it comes to the DAC routing and contain at least a couple of bugs (the biggest one being that they reversed the entire bit order, so it looked like output 16 from the LA chip was routed to bit 1 on the DAC, output 15 to bit 2, etc.). If the schematic is to be trusted, then not only is bit 16 on the DAC grounded, but output 16 is routed to bit 15, and output 15 is skipped entirely (not routed anywhere). That is:

Code:

LA/Reverb bus|DAC
   1         |  1
  ...        |...
  14         | 14
  15         |gnd
  16         | 15
  ground     | 16

However, I have serious doubts about the accuracy of the schematic. Would be best to have it confirmed by someone with an actual unit.

I'm now wondering how much cleaner this 01 board sounds, and thinking that the noise level is probably very close to, if not exactly the same as that of the CM-series.

Too late for me to check the rev 01 board now. Will do so tomorrow.

There is a guy on eBay claiming that the old-revision MT-32s without the headphone jack are superior because they use a "faster processor." True or false?

Uh, off-hand I'd say the statement was false, if only for the fact that superiour is so vague a definition as to make it useless.

[Cloudschatze]

There is a guy on eBay claiming that the old-revision MT-32s without the headphone jack are superior because they use a "faster processor."  If so, this could be another reason why some MT-32 songs don't playback correctly on later modules if they also use a slower processor.

According to Martin Sant's information, the revision 00 board has an Intel 8097 MCU. The revision 01 MT-32, and CM-variants contain an 8098 MCU.



Which songs will not play correctly on the later modules?

[Great Hierophant]

http://cgi.ebay.com/Classic-Roland-MT-32-Module-MT32-LA-Synth-CLEAN_W0QQitemZ7343744061QQcategoryZ38090QQssPageNameZWDVWQQrdZ1QQcmdZViewItem

is the auction I was referring to.  I have found a datasheet for the 8097MCS and the 80C198 used in the CM-500.  The specs for the latter chip equal or beat the specs of the former chip, so unless he is referring to the LA or Reverb processors he must be talking out his ass.

[Laust]

[...]he must be talking out his ass.

I think that sums it up very nicely

[Great Hierophant]

I believe that the more common revision 00 boards may be better suited to playback of music composed for an MT-32, the noisier output being a defining characteristic of the module.  Less noisier revision 01 MT-32s seem to equal CM-32Ls or LAPC-Is without the extra sound effects.  I would get a revision 00 for MT-32 optimized music and keep a spare CM module for the CM specific stuff.

[NewRisingSUn]

I would get a revision 00 for MT-32 optimized music

I personally prefer the cleaner sound of the CM modules. In fact, I'm currently thinking about setting up a way to grab the digital sound from the mainboard itself before it even reaches the DAC, instead of recording the analog output, giving me the purest LA sound conceivable AND preserving the integrity of the digital chain (resulting in true DDD CDs).

As far as "optimization" goes, it's hard to tell what a game is "optimized" for. Most game composers, apart from Sierra that is, used a LAPC-I because that's what most gamers had. And even if you know from the composer himself that it was an MT-32, you still don't know if it was a revision 00 (and most likely, the composer doesn't know either). Just look at how complicated it was to find out that Larry 5 actually was composed with a CM/LAPC.

But if what we learned in this thread is true, you can get the revision 00 noisiness from a CM-32L/revision 01 recording by just substituting the lowest bit of every sample in the recording with 1-bit noise.

I find it actually very interesting that people would find the revision 00 to sound "warmer" --- noise resulting from inadequate digital processes and especially from bad D/A conversion is usually is perceived as "cold" and "lifeless" (like early CD palyers), not "warm".

[Alistair]

I find it actually very interesting that people would find the revision 00 to sound "warmer" --- noise resulting from inadequate digital processes and especially from bad D/A conversion is usually is perceived as "cold" and "lifeless" (like early CD palyers), not "warm".

I think that's a good point. I think I understand why though- it all depends on your digital playback equipment.

Like, I run some 3 Sound Blaster cards (AWE32, SB Live!, SB Audigy 2) and the MT-32 sounds lovely with that 'warm' sound. I assume that's to do with the bass and treble.

My better quality equipment, my new TB Santa Cruz and my M-Audio FireWire Audiophile both sound starkly different (in terms of recording WAV's and mixing Sierra CD's). All instruments are much more 'well-defined'. I find SB output 'blurry', because of that warmness.
I find the Santa Cruz middle-o-the-road, and my expensive Audiophile's high quality means I can hear MT-32 hiss and the coldness you describe, NRS, how it really sounds.

I assume therefore that the majority of folks run a Sound Blaster digital card and thusly the D->A and A->D conversion is done worse, or the cards just sound different because of their different processing, or something. (I get the ideas, but not the specifics, if you know what I mean..)

Of course, I could be talking out my ass. That's just what my ears tell me from detailed listening with all MIDI devices and sound cards.

- Alistair

[Great Hierophant]

As far as "optimization" goes, it's hard to tell what a game is "optimized" for. Most game composers, apart from Sierra that is, used a LAPC-I because that's what most gamers had. And even if you know from the composer himself that it was an MT-32, you still don't know if it was a revision 00 (and most likely, the composer doesn't know either). Just look at how complicated it was to find out that Larry 5 actually was composed with a CM/LAPC.

I understand the difficulty, so I advocate getting both.  As the LAPC-I and CM-32L were released in March and April 1989 or thereabouts, and given the time it took for those devices to penetrate to the market I would suggest that only VGA games would have been composed for LAPC-I or CM-32Ls. Additionally, I hear that revision 0 LAPC-Is had the same noise issue that the early MT-32s had.  

Leisure Suit Larry 5 is among the last games Sierra had designated the MT-32 as the optimal music device.  Somehow I doubt it is the only late Sierra title that supported the LAPC-I/CM-32L specific sound effects.  Of course, when King's Quest VI came out supporting General Midi/GS devices for the best music, it didn't much matter anymore.

[Alistair]

Leisure Suit Larry 5 is among the last games Sierra had designated the MT-32 as the optimal music device. Somehow I doubt it is the only late Sierra title that supported the LAPC-I/CM-32L specific sound effects. Of course, when King's Quest VI came out supporting General Midi/GS devices for the best music, it didn't much matter anymore.

Well, assuming the premise about LSL5 is correct, I'd say it's because Safan composed it, as opposed to a more regular 'in-house' Sierra composer.

I guess PQ3 and PQ1VGA could have also been composed for the CM (PQ1VGA doesn't quite sound right on the MT-32 sometimes, even though i's composed for it, as far as I'm aware, and I bring up PQ3 because it was Jan Hammer). But not having one, I wouldn't know. .anyone else know of any others?

- Alistair