New AOUT module design?

[TK.]

Hi,

in the meantime there are so many people with hardware design skills here in the forum, that I would like to ask if somebody is interested in creating a successor of the AOUT module which is not based on the expensive MAX525, and which is easier and faster to build than a AOUT_LC module.

With design I mean to develop the circuit, to create the PCB layout and to write some documentation

Only requirement: the used chip(s) should be easily available.

Number of channels and gates + additional options (e.g. CV offset, etc.) to be discussed w/ the community

I will take over the software part if required...

Best Regards, Thorsten.

[stryd_one]

I am a bit busy at the moment but I'll volunteer any help I can give.

[seppoman]

Hi,

I´ve been searching for nice DACs in the last time. A very cool IC is the TLV5630 from Texas Instruments. It is a 12 bit, 8 output converter with internal reference (no MAX6007 anymore) and a cascadable SPI interface. Only drawback in comparison to the max525 is that it doesn´t have a gate logic output. But this could be either solved via the normal DOUT chain or a dedicated 74hc595 integrated to the AOUT board. It is available from e.g. Farnell for 15.49 Euros including VAT (1-9 pieces). If the real El Cheapos want to save even more, there are also pin- and instruction-compatible 10 bit (TLV5631) and 8-bit (TLV5632) versions available. And TI´s free sample program is is also really world class compared to Microchip/Maxim etc.

This IC is only available in SOIC form factor, but I think people building an AOUT are more ambituous than average anyway and the OPL3 board has already shown to a number of MidiBoxers that there´s nothing to be afraid of SOIC SMDs as long as you have a premade PCB. And for 8 (!) AOUT_LC DAC channels you´d need to solder hundreds of resistors, so one SOIC is definitely the better choice.

Honestly, I don´t have much experience in DAC design up to now and I don´t have any CV gear around to do real world testing. But I´m in an extensive layouting phase anyway so Thorsten, if you have a look over the data sheet to confirm that this DAC is suitable and major changes in the OpAmp part of the AOUT won´t be necessary, I could do a board. I´ve got TLV5630s lying around here and also have the opportunity to etch boards at university, so after a few questions are settled I could send you a free completed board for testing and programming :)

Seppoman

[TK.]

Hi Seppoman,

from the feature list and from the availability (it's even available at Mouser) it looks very good! :)

Only disadvantage I see is the high linearity error of +/-2LSB typical and +/-6LSB max (the MAX525 has +/- 0.5 LSB), which means +/- 3 mV, but I think that this is acceptable (or do I oversee something?)

SMD is ok for such a module (total newbies should start with the R-2R latter, this gives them a training on handling the soldering iron ;-))

8 trigger outputs (-> 74HC595) are a must

Best Regards, Thorsten.

[seppoman]

Hi Thorsten,

Only disadvantage I see is the high linearity error of +/-2LSB typical and +/-6LSB max (the MAX525 has +/- 0.5 LSB), which means +/- 3 mV, but I think that this is acceptable (or do I oversee something?)

yes, I also noticed that this afternoon, and it gave me a bit of a headache. Actually, if my calculator doesn´t lie, +/-6 LSB is around 15 mV? if full scale is 10.67 V (1V/Oct) then 2 LSB is 5.2 mV and 6 LSB would be 15.6 mV. 1 Semitone is 83.4 mV, so 2 LSB is already more than 6 cents detune and 6 LSB is inacceptable!! Still an error of 6 LSB is even better than an 8 bit DAC... I wasn´t aware that CV is so receptible to tuning errors and you can´t get a single clean interval off an 8 bit CV system.

Even though the curve in the datasheet shows "only" INL of around +2/-1 LSB I´d still say this DAC would be noticeably inferior to the max525.

But TI has some other nice DACs. Sadly there are no better 8 output DACs in usable packages, but there´s the DAC7614UB (TI/BurrBrown) with a max nonlinearity of 1 LSB. The curves in the datasheet show that it´s typically around +/- 0.25 LSB! It needs an external reference but should be quite simple otherwise. It is 12.60 Euro at farnell for SOIC. There´s also a PDIP version but the equivalent PB type would cost 25 Euros, so I´d go for SOIC anyway. It´s also available from Mouser.

What do you think?

Seppoman

[seppoman]

Hi,

just wanted to add new findings :) :

There´s another 8 channel 12 Bit SPI DAC I´ve found: The Linear Technology LTC 2620. It has INL specced 0.75 typical and 4 max. It is available from RS components for around 18 Euros. Of course the INL is not as good as the DAC7614, but it´s 8 channels, has less unnecessary control pins and at least the specs are noticeably better than the TLV dac. I don´t know how probable it is that the INL goes higher than the typical value and under which circumstances this can happen? Is it just that rarely you can have bad luck and get a bad one or does the value rise e.g. with temperature or age? Another problem is the SSOP16 package (Pin spacing is 0.65 mm). I would be able to solder it but it´s even less newbie friendly than SOIC. So all in all I´m not too enthusiastic about it...

BTW, there is also a DAC7615 version that is double buffered (otherwise same specs like 7614). I don´t know if this is necessary or useful and I haven´t found it under around 20 Euros.

regards,

Seppoman

[stryd_one]

Thanks for taking the time Seppo :D

I know it can take many hours just to find a single candidate, and sometimes it falls through right away. You must be losing sleep by now ;)

[moebius]

Hi TK, Seppoman.

Getting the picture? ;D

I spent numerous hours searching for acceptable replacement for MAX525s, although I had a bit narrower search criterias. I really couldn't find anything even nearly matching the needs.

I guess price/performance ratio in 525 is about the right? In a way, I'm questioning, if this is really worth the effort?

For the most MAX525s are one time investment, off course an expensive one when compared to other MIDIbox related stuff. But usually then, the stuff to be controlled is expensive too?

Would the concept of using shift registers and parallel interfaced dacs make the search for a proper and cheap dacs easier?

Moebius

[TK.]

These a good points, Moebius!

In general I'm not sure if the price/quality ratio of the MAX525 is still justifiable today (2 years have passed after the first release), and on the other hand I see users (including me) who just only want to control some CC's (like cutoff, resonance via MBSID or MBFM), where linearity doesn't really matter, and R-2R latters consume too much area if you want to realize more than 2 channels w/ high resolutions.

But your point about parallel interfaced DACs should also be considered - are there better/cheaper ones available (I don't think so, but I could be wrong).

Best Regards, Thorsten.

[seppoman]

Hi,

another day of searching - Maxim doesn´t offer anything at a realistic price. Analog Devices are either expensive, too, or they come in extremely small packages.

I also searched for parallel DACs but most have only one or two channels and are more expensive than the serial versions anyway. And with shift registers, few channels per chip mean a lot of shift registers (if there are dedicated data lines for every DAC) or an awful lot of bridges on the board. I´d say with the 80x100 mm restriction it would be even hard to get 4 channels onto one board. One of the goals of a redesign should be simplicity, and that is contradicted by the shift register concept.

About the DAC7614/15: I think it has the best value/price ratio. The specs are really good. The only real disadvantage is that it has no cascading output, i.e. probably we´d have to use two outputs of a 74hc595 to control the CS and LOADDAC lines.

If using this DAC, perhaps it would be a good idea to put only 4 channels (one DAC) onto one AOUT board. This way people needing only few channels for filter control could save half the money and effort.

Thorsten, could you please comment on the DAC7614/15 and the LTC2620?

Seppoman

[TK.]

I don't like both of them. The serial protocol of the DAC7614 (especially the LOADDAC# line) will cost too much performance when many outputs should be updated periodically (MBFM, MBSID), and the LTC2620 comes only in a Lead Package, which means, it could happen, that they won't be available in 1-2 years anymore.

A quad-DAC on a single board would be ok. Something similar to the DAC7614 (it can optionally be supplied with +/- 5 V, this is really great!), but with serial output and w/o the need for latching the DAC address would be perfect

Best Regards, Thorsten.

[seppoman]

Hi Thorsten,

the LTC2620 comes only in a Lead Package, which means, it could happen, that they won't be available in 1-2 years anymore.

why´s that? Is there something about the SSOP package that is announced to be discontinued or did you confuse "16-Lead package" with leaded? The part is available RoHS conform.

Something similar to the DAC7614 (it can optionally be supplied with +/- 5 V, this is really great!), but with serial output and w/o the need for latching the DAC address would be perfect

Hmm, I´ll continue to search, but I think I´ve seen most of the established companies parts now, and the DAC probably shouldn´t be too exotic.

Another idea: Coming back to the shift register idea - what about using a single parallel DAC in combination with 8 S&H ICs. The LF398 looks nice and is around 2 Euro. And there are parallel versions of the 7614. The 7624 is quad (but also with loaddac# line etc), but there´s the 7613 single dac. With only one dac you could probably keep the C/S, LOADDAC and R/W at a fixed state (as the 595s are latched at once anyway) and just set the s/h control alternating. Only problem is that the 7613 is also only available as SSOP package.

Seppoman

[TK.]

Sorry, maybe I misunderstood the term "Lead package" - I noticed via findchips.com, that the chip is not available in such a high quantity like the other ones, therefore it looks like it is discontinued. But I could also be wrong...

Also S&H is very bad for the performance. Especially for MBSID and MBFM, where I want to control multiple external filters without loading the CPU that much. Under these circumstances I would prefer the TLV5630 before the others, since a LSB error doesn't matter so much here.

Really difficult decition! Can't understand, why manufacturers don't use such a perfect interface/protocol like the MAX525

Best Regards, Thorsten.

[adamstan]

Don't know if it's right place to post, but...

I've designed a replacement for the AOUT_LC module, using two 4-channel 8-bit AD7226 DACs. They have parallel interface, so I used two shift registers (one per chip) and 2051 for addresing. It is supposed to work like this (I've made some minor changes to cv_aout_lc.inc):

- Two shift registers are loaded

- Values are latched into registers, and in the same tame, latch signal triggers interrupt on 2051

- 2051 latches values into DAC by setting !WR pin HIGH

- 2051 increments DAC address

- 2051 sets !WR LOW, enabling new voltage to be written in next channel

- next two bytes are loaded into SR's... and so on

But after what I've read in this topic, I've checked its accuracy, and with 8bits, +-LSB or +-2LSB is really bad for driving V/oct VCO in polysynth... :-/

Am I doomed to buy MAX525 or clone some 80's circuits with single 14bit DAC and S/H? Or are there other ways?

[DrBunsen]

Perhaps it would be worth asking Doepfer what DAC they are using in this new prototype.  Not that Doepfer are likely to be using "cheap" components, but they must have checked to make sure of continued supply, and accuracy.

[e-toy]

I also have searched houres to find a new DAC type, without a real hit.

no usable DAC with Serial input, nor Parallel.

but, what about a I²C DAC ?

maybe like the DAC7574 or LTC2625 ? Not the best technical spec, but maybe better to implement ?

e-toy

[seppoman]

Hi,

I2C is a relatively slow protocol, so I don´t think it would be a good idea to use I2C DACs.

Anyway, just to keep you updated: Thorsten and I had a talk about this issue in December and decision was made that I will design a new AOUT based on the TLV5630 in near future. The TLV is a reasonable compromise between cost, no of outputs and precision. It probably won´t replace the MAX525 AOUT, but is a cheap and easy to source alternative for people who want to drive e.g. a number of filters with it. The precision will at least be better than an AOUT_LC (without doing really extensive resistor matching in 12 bit mode), and you get 8 channels for only 15 Euros for the DAC.

At the moment I´m really busy with boards for another project, but I will start the design in a few weeks and keep you posted :)

Seppoman

[MyCo]

You could replace the MAX525 with a 74HC595 and a DAC 08 (or DAC 0800, DAC 0832) total costs for this would be 3 Euro (eg. @reichelt). Accuracy would be +/-1 LSB and with a DAC 0830 LCN you can get a linearity error less then 0.1% of the fullscale which would be 10mV...

with an additional lowpass filter and a good opamp eg. OP 27 it would be a perfect replacement

[seppoman]

Hi,

This would be a nice idea for replacing the AOUT_LC. But one 595 and one DAC08(00) are still only one channel. For 8 channels it wouldn´t be cheaper than the TLV5630 and the board would get quite crowded and/or big. A lot of soldering to do, too. And +/- 2 LSB typical of a 12 Bit converter is better than +/- 1 LSB of an 8 bit converter.

Seppoman

[e-toy]

thank you guys for discussing

I'm a noob in digital electronis, but I like to learn to understand it.

@seppoman

your project is very interesting as cheap alternative to the AOUT (8 channels)

I'm looking forward and hope you will find the time to realize this project.

@ MyCo

You're right. But this is nearly the same as the AOUT_LC. Exept that you change the resistor ladder with the DAC. It shurley looks better, may it's easier to build and hopefully works more precise then the AOUT_LC version.

Did you have made such a AOUT module ? What about the driver (and schematic) ?  :)

@all

what's the benefit of 12 bit DAC's for CV controlled By CC (7 Bit) ?

I think there should be a interpolation of CC data (SW) to obtain a finer stepping of CV,

or the CV will allways have only 127 steps independent if you use a 8 bit or 12 bit DAC.

or am I wrong ?

e-toy

[seppoman]

@ MyCo

You're right. But this is nearly the same as the AOUT_LC. Exept that you change the resistor ladder with the DAC. It shurley looks better, may it's easier to build and hopefully works more precise then the AOUT_LC version.

Did you have made such a AOUT module ? What about the driver (and schematic) ?  :)

At least in 8+8 bit mode, the normal AOUT_LC driver should work with such a module. It doesn´t matter if there´s a resistor ladder or an 8 bit Dac behind the 595s.

@all

what's the benefit of 12 bit DAC's for CV controlled By CC (7 Bit) ?

with CC control, of course the resolution is limited. But as far as I know the filter and envelope curves of MBFM and MBSID are already interpolated internally. And especially when putting out not only control data but VCO pitch voltage, the resolution has enormous impact on the tuning precision of the VCO.

Seppoman

[DrBunsen]

I2C is a relatively slow protocol, so I don´t think it would be a good idea to use I2C DACs.

How about CAN?  Thorsten's just posted that he's evaluating a CAN capable PIC.

TLV5630

Do you have a link for this chip?

8 channels for only 15 Euros for the DAC.

And it's 12 bit +/- 2 bit?  Woohoo!

[DrBunsen]

Just discovered the CANANA project:  PDF  HTML (mangled by Google)

The CANANA outputs 16 analog 14-bit voltages (0-10Volt) and inputs 16 analog 16-bit voltages (0-10 Volt)

...

A 14-bit SPI DAC (AD5390) is in charge of the 16 analog outputs.

A 16-bit SPI ADC (AD977) is used for reading the 16 differential inputs.

...

DAC/ADC offset and gain errors (including the analog chain errors) are within +/- 2 LSB after calibration.

...

The ADC and the DAC use the same external high precision 3 ppm/C reference (AD780).

...

The DAC is a 14-bit device, with a 610 microVolt LSB. The settling time of the outputs is around 100 microseconds, for a 10 Volt step. A complete output setting will thus take 230 microseconds (one CAN frame plus one conversion). The DAC has a guaranteed relative accuracy of +/- 3 LSB, and neglictible drift. The maximum DAC output voltage error will be +/- 2 mV.

[seppoman]

How about CAN?  Thorsten's just posted that he's evaluating a CAN capable PIC.

I didn´t see any DACs with CAN interface. Do you know one? I think CAN is meant to be an inter system bus and not a way to connect peripheral ICs.

Do you have a link for this chip?

Google is your friend - first hit:

http://focus.ti.com/docs/prod/folders/print/tlv5630.html

And it's 12 bit +/- 2 bit?  Woohoo!

it´s 12 bit +2/-1 LSB INL typical, max 6 LSB. The TLV is not quite as good as the MAX525, but 2 MAXs are around 60 EUR and hard to get anyway. The module won´t replace the old AOUT for all purposes.

About CANANA: This is really a whole system with own processor that is communicating via CAN. Doesn´t seem as if you can buy it, and it would be quite expensive.

But the AD539x series has really great specs - the 14-Bit DACs have around +/-1 LSB INL (!) - that´s even better than the 525 :) It has a cascadeable serial interface, too. The only problem would be the LQFP package with 0.65 mm pin distance  :-\ 16 channels would be a bit overkill and if all 16 channels get their OpAmps etc., everything wouldn´t fit onto a 80x100 mm board. But there´s the AD5392 8 channel version that is available at Farnell for 33 EUR +VAT. So maybe this one is worth some discussion - Thorsten??? :)

Seppoman

[cheater]

Guys,

Maybe I'm very wrong here, but with small typical error and a big maximum error, if you output the same value many times you will get the "typical" range most of the time. Does AOUT only output (and hold) a single value? Or if you hold a certain note will it keep on outputting that certain voltage (and each time with a different error - the longer you output it over and over the more of this "typical" range you get there)?

Maybe a circuit which takes in around 200 values and rounds them all off would be simple to make. You don't need the CV outputs at 192 kHz and stuff! So using this extra precision this way could be feasible.

Thanks,

D.