DragonMaster Posted April 21, 2008 Report Posted April 21, 2008 Well, the MT ones seem to have good crosstalk specs when you keep the impedance low. Also, the X and Y lines are I/Os, there's no assigned input and output pin. They say you can connect the crosspoints in any way you want. It seems more flexible than the 8113. Also, it's not only available as an LQFP package. The only disadvantage I see is an higher THD.How do you handle summing though if you use relays? Wouldn't a transistor/op-amp based approach be better?Buffers on the inputs and outputs? Yikes! That starts to be complicated, and you need components around the relays to prevent pops...Otherwise, I guess one could use a 100kOhm resistor grid with "muting transistors" between every crosspoints. Quote
Screaming_Rabbit Posted April 21, 2008 Report Posted April 21, 2008 How do you handle summing though if you use relays? Wouldn't a transistor/op-amp based approach be better?... what now? A Patcher or a Mixer ;) What benefit is expected (thinking in working process) other than Stereo to Mono summing?Greets, Roger Quote
DrBunsen Posted April 21, 2008 Report Posted April 21, 2008 Hmm, good question. I'm going to have to think a bit more clearly about that. /edit/ but the OP did request itOutputs should also have the option of summoning/combining any number of the input signals / But the input should stay available to other outputs / I guess they would all need an individual gain/level options or something anyone seen Nomical?Hmmm, a 16x16 mixer on an IC sounds like a big call. And 256 discrete digipots with addressing ..... :PAs an aside, I should mention that I'm looking at something that can route CV and/or audio.Oh and TK you big tease:Your search - docmatrix midi router - did not match any documents. Quote
DrBunsen Posted April 21, 2008 Report Posted April 21, 2008 Ken Stone has some thoughts:Cascade MixerC.V./Audio Mega MixerD.C. MixerDev-mod Mixer/Inverter block5 x 5 Matrix Mixer Quote
stryd_one Posted April 22, 2008 Report Posted April 22, 2008 Well, the MT ones seem to have good crosstalk specs ...The only disadvantage I see is an higher THD.I'm glad you're looking at these specs. Something worth considering:Crosstalk does not cause such problems when you expand the matrix by chaining IC's, because the neighbouring channels change... Distortion, however, cascades and increases with every IC it passes through, as does noise. "muting transistors" between every crosspoints.That's an interesting idea I haven't heard thrown about before.... Quote
/tilted/ Posted April 22, 2008 Report Posted April 22, 2008 But surely to expand a crosspoint matrix, you would parallel the chips, not chain them?Ie (using 2x4x4 for clarity)[tt] Y1 Y2 Y3 Y4 | | | |X8 -+---+---+---+- | | | |X7 -+---+---+---+- | | | |X6 -+---+---+---+- | | | |X5 -+---+---+---+- | | | | Y1 Y2 Y3 Y4 | | | |X4 -+---+---+---+- | | | |X3 -+---+---+---+- | | | |X2 -+---+---+---+- | | | |X1 -+---+---+---+- | | | |[/tt]I don't see why a chip manufacturer would bother to include bus outputs as well as inputs, as these would simply carry the same signal as the input, plus a wee bit of distortion and noise.By adding chips, you are adding to your noise floor, but not neccessarily adding to distortion, as your crosspoint is only happening on one chip. Quote
stryd_one Posted April 23, 2008 Report Posted April 23, 2008 But surely to expand a crosspoint matrix, you would parallel the chips, not chain them?Think bigger dude. Think 64x64. Think 128x64. Think 256x256 (I wish). Now you're thinking in my league. 256x256 would just take care of my requirements but would not leave much room for future growth. You don't just add them 'vertically' as you have, but also horizontally. Sure, not everyone wants one as big as that, but it's better to aim for the top of the mountain and get 2/3 of the way up, than to aim half way and make it ;)This is also the reason why 16x16 IC's are such a big deal. Sure, you can use 8x8 chips in a matrix to do the same thing, but.... Quality takes a dip. Quote
Screaming_Rabbit Posted April 23, 2008 Report Posted April 23, 2008 Think bigger dude. http://www.studer.ch/index.aspx?menu_id=3&sub_menu_id=12&locale=en&url=includes%2fproduct_sheet_include.aspx%3fproduct_id%3d43Now you're thinking in my league 8)Greets, Roger Quote
stryd_one Posted April 23, 2008 Report Posted April 23, 2008 Hahahah yeh there are a few big commercial offerings out there... Seen the pricing on those though? Ouch!More reasons to love midibox :) Quote
/tilted/ Posted April 23, 2008 Report Posted April 23, 2008 Think bigger dude.I think you're missing my point.Doing a 4x4 matrix would be close-to useless, but drawing more than a 4x4 would have made my point less clear.Ie (using 2x4x4 for clarity)The point again:But surely to expand a crosspoint matrix, you would parallel the chips, not chain them?This was in response to:Crosstalk does not cause such problems when you expand the matrix by chaining IC's, because the neighbouring channels change... Distortion, however, cascades and increases with every IC it passes through, as does noise. What I'm saying is that whatever the size of your matrix, you are still routing one-to-N, not N-to-one *. To do N-to-one would still require a summing mixer, or horrible gain problems will follow.This means that your output is derived from only one crosspoint, on one chip.You don't gain anything by chaining chips.Noise will increase as you add chips, because you are connecting outputs together. But not distortion. You are not running through one chip, into another. Doing this does not increase the size of your crosspoint matrix. It simply makes your path more convoluted. ;)*Where N is the set of numbers >=0... This could include 16, 256 or 512 gujillion. :P Quote
stryd_one Posted April 23, 2008 Report Posted April 23, 2008 I know that you used a scaled-down example and understand why :) Damn if I expect you to quadruple the size of your ASCII art !So yeh, no need for the thrice-over drum-in. That's just rude :PFunnily enough your ascii art didn't quite explain it to me but when I went to the schems as a reference I saw the 4x4 chip layout and suddenly 'got' it - as well as the matrix outputs being combined like you drew, each input is split into X amount of busses going into each chip horizontally - and suddenly I understand the need for the input buffers! Yay! And less stress about specs! Yay more! Thanks for taking the time to explain it to me. It's only because of dudes like you doing stuff like that, that I was able to understand the schem this time around. I guess I'm learning something about all the wires behind these screens after all :DWhat I'm saying is that whatever the size of your matrix, you are still routing one-to-N, not N-to-one *. To do N-to-one would still require a summing mixer, or horrible gain problems will follow.Totally. Where'd you get that from? Remember me saying to bunsen that he was after a mixer? ;) Of course, maybe you could always just reduce the input levels at the source (turn your synth down) .... But anyway I wasn't talking about doing that, let's move on :)Thanks! Quote
gavgomad Posted April 23, 2008 Report Posted April 23, 2008 Well, maybe this makes the Mitel chips a bit more attractive? I know the 8816 is a little odd in that it is an 8x16 (rather than an 8x8), but at $4.50 US each, you could chain quite a few together in no time for a lot less than the $40+ US each of the 16x16 chips....I'd be happy with a 32x32 crosspoint myself.... Plenty for controlling the instruments going to the board, and switching some key in-line effects in and out.... so theoretically 4 16x16 chips would do the job, but that's still a steep request for samples from AD, no?Just thinking aloud.... ;) Quote
stryd_one Posted April 23, 2008 Report Posted April 23, 2008 Well, maybe this makes the Mitel chips a bit more attractive? I know the 8816 is a little odd in that it is an 8x16 (rather than an 8x8), but at $4.50 US each, you could chain quite a few together in no time for a lot less than the $40+ US each of the 16x16 chips....Zactly!!Being 8x16 is no issue, given what tilt has demonstrated above...What is the cost of the buffers likely to add up to for that though? If you include the cost of the 32 opamps and the extra board space etc, how does it compare to the AD chip (which has the buffers on-chip)? Quote
gavgomad Posted April 23, 2008 Report Posted April 23, 2008 What is the cost of the buffers likely to add up to for that though? If you include the cost of the 32 opamps and the extra board space etc, how does it compare to the AD chip (which has the buffers on-chip)?Well, I guess that depends on how boutique you want to get with your opamps? I would think 8x TL-074 low noise quads should be sufficient for the job? Again, just using Futurlec as a price example (not saying it's the be-all and end-all place to buy op amps!), at $0.40 per DIP TL074s ($0.20 if we go surface mount), the savings is still pretty hefty.... Considering there is almost no board space for the actual MT chips, 8 quad buffers wouldn't be too out of line in terms of board size?Gav. Quote
stryd_one Posted April 23, 2008 Report Posted April 23, 2008 Boutique is for fairies. I would want to go with a nice sounding option, but I imagine it's quite possible to put the buffers on a separate board so that people can pick and choose... TL074 seem to be the run-of-the-mill, I guess that's for a good reason... But I magine that perhaps with a little extra you could roll something really nice into it... But I wouldn't know a good one from a crap one until I heard it... Anyone here an opamp guru? Quote
gavgomad Posted April 23, 2008 Report Posted April 23, 2008 Hey Stryd,I'm no guru by any stretch.... But I note in the MIDIBox Mixer project there seems to have been a consensus to go with the OPAx130 series? A little more "boutiquey" than an 074 (probably more expensive as it has that "Burr-Brown" badge of honour). BUT, in the DIP version of OPA4130 (quad) the pinout is the same as the 074, so I guess barring any instability in the OPAs requiring caps to prevent oscillation or what not, you could use the same layout and try either / or depending on your tastes?Remember, many commercial products use lowly 4558s for both amplification and buffer duties, so an 074 HAS to be better than those! ;P Quote
gavgomad Posted April 23, 2008 Report Posted April 23, 2008 OK, my brain is starting to fail me here.... IF we go with the MT8816, I'm just trying to figure exactly how many you would need to get a true 32x32 crosspoint? The diagrams I have seen seem to use both of the 8x inputs (lets call that our "X") to interconnect pairs of MT8816s, but this theoretically limits the number of connections in the matrix to 8?The 16x16 chips are easy - 4 chips and you're there.... How many MT8816s would it take to do the job PROPERLY for a 32x32 crosspoint (ie. 1024 switches all addressable)?(the hamster wheel in my head is a little squeaky these days! ;-P)Gav. Quote
/tilted/ Posted April 23, 2008 Report Posted April 23, 2008 So yeh, no need for the thrice-over drum-in. That's just rude :PSorry, didn't mean to be rude.I guess there's ramming home the point, then there's being a {expletive deleted}.Damn I miss the chat. :'(Boutique is for fairies. Hear Hear! A large number of op-amp designs use same pin outs. I think for a time there was a bit of "me too", "mine's the best for everything" going on in the op-amp world, and the manufacturers used same pinouts to ease prototyping headaches.I think the idea of using a seperate buffer board is a good call, that way if some crazy audiophile or guitar hero wants to use discrete OTAs, germanium transistors, NOS radio valves from a soviet submarine pre-amp, they can.Possibly, we could also modular-ise the matrix board, say putting a 16x8, or 2 16x8s on one board, with input thrus and output nodes to connect more matrixes as needed. Make the board to get around the one-set-of-inputs, one-set-of-outputs thing. Quote
DrBunsen Posted April 24, 2008 Report Posted April 24, 2008 Well, first thing I was planning to do was to mount the SMT AD chip onto a small PCB with 0.1" pins underneath, so I can plug it into different prototype carrier boards without mucking about with surface mount soldering. Quote
stryd_one Posted April 24, 2008 Report Posted April 24, 2008 OK, my brain is starting to fail me here.... IF we go with the MT8816, I'm just trying to figure exactly how many you would need to get a true 32x32 crosspoint? The diagrams I have seen seem to use both of the 8x inputs (lets call that our "X") to interconnect pairs of MT8816s, but this theoretically limits the number of connections in the matrix to 8?You've been confused by the same confusing diagrams I was confused by. How confusing.It's 8 chips: Inputs 1-8 9-16 17-24 25-32 /\ /\ /\ /\ | | | | | | | | | #---- | #---- | #---- | #------- 1 -16 Outputs | | | | #--------#--------#--------#---------- 17-32 Each set of 8 inputs is split in two and goes to two chips in parallel, so that each set of 8 inputs is capable of being routed to any one of the 32 outputs, which are in two groups of 16. Possibly, we could also modular-ise the matrix board, say putting a 16x8, or 2 16x8s on one board, with input thrus and output nodes to connect more matrixes as needed. Make the board to get around the one-set-of-inputs, one-set-of-outputs thing.Hehehethinking about connections to chained boards for different architectures (16x32, 32x32, etc) and separate interface boards (balanced, unbalanced, etc)Question: When you do a balanced IO I know you have to use an opamp to do the cancelling thingy (I think I understand what it does with flipping the noise to remove it and all but I don't know the electronic terminology properly ;) ) does that become the buffer also, or is that additional?Tilt: Hah! Don't sweat it bro! :) I noticed in my headphone-amp travels that lots of opamps share pinouts so I guess that puts that to bed... I reckon base it on the 074 and leave it to the builder to fiddle around :) First one to try a discrete opamp shall forever be known as "Tinkerbell".It's a tough call about one or two chips on board... Or in fact three. 8 into 16 is probably a bit small to be useful for studio routing, mest synths have more IO than that and are internally routed. However it could be cool for modular analog synth patch storage (yaknow, saved patches on your big moog). Could also be cool for FM or the SID. Of course you'd have 8 outputs spare, but you could route the audio back around into the AINs or the SID inputs and... oh don't get me started...16 into 16 is probably about the right size for most hobby studio usage. DOCmatrix is clever-clogs and does 24x24 which is smart because almost all patchbays are 48-point (2 rows of 24 each side, total 96 counting front and back), and that's a relatively cheap way to get an enclosure and jacks, all pre drilled etc...My personal tilt (sic) on this:48x48 would be a great size for me because i can use my existing patchbays which will of course be rendered obsolete (yay, die you manually controlled scum, die.). That's 6x3 chips, which could be 3*boards with 2x1 chip. So, I'm biased toward the 16x16 layout. Also I think that the extra IO could be inventively used if it were for a synth. Only bummer is, it doesn't allow for that 24x24 which people might like ...I get a cool "go team" vibe off this thread, nice one dudes. Quote
/tilted/ Posted April 25, 2008 Report Posted April 25, 2008 ...so, which chip?do we have a concensus? Quote
stryd_one Posted April 25, 2008 Report Posted April 25, 2008 OK you made me read the datasheets again :)I thought the 8816 was the way to go, although compared with the AD8113 the distortion figures don't look so hot - 0.12% at 20khz for the zarlink, vs 0.002% for the AD. That's a whopping difference :(You'll need the appendix to compare the figures equally, cause the zarlink site and datasheet give specs at 1khz.MT8806/08/09/12/14/15/16 Analog Switch Array Measurements Quote
/tilted/ Posted April 25, 2008 Report Posted April 25, 2008 The AD8113 does look very nice.It also seems to be one of those classics from the audio realm, of "for something a little better, insert all of your money here"Something like $30 US per chip...Also it is only in a 100pin LQFP. Are we OK with that? Quote
stryd_one Posted April 25, 2008 Report Posted April 25, 2008 Yeh it's about 30-35... honest such a high distortion figure on the zarlinks kinda makes me feel it's worth it.... But as I said I want it to be good for everyone, and making it too expensive would suck.Gav, where did you find that $4.50 price for the MT8816? I couldn't even find somewhere to buy them! Anyway assuming that's correct, if you consider two chips plus buffers plus board space for them, it comes up to $15.. still a lot cheaper...Well, as DrBunsen pointed out the 75019 might be available again soon. Actually the JP should be today (the date didn't get pushed out!)... It's PLCC with those J legs and you can get/should use sockets for them so no need for SMD soldering. Chip's $17 (by the 100 from AD), but again we'd need the buffers so add $5. Both crosstalk and THD is similar to the 8813, and buffers add so little distortion it's hardly worth considering. Seems like it might be a good middle-of-the-road. Upside for some of you is that it's a bare-crosspoint so there's no fixed inputs and outputs. Also there's some work been done by others from which we can plagiarise take inspiration ;) http://www.circuitcellar.com/library/print/hcs-pdf/45-Ciarcia.pdfhttp://synth.stromeko.net/DIY.html#Matrix32x32We're spoiled for choice I guess ;) Quote
gavgomad Posted April 25, 2008 Report Posted April 25, 2008 Hey all,The MT8816 is available at:http://www.futurlec.comJust type MT8816 into the search bar.... $4.50 a chip is pretty tough to beat, particularly with how many we may need.... ;-PThat being said, I hear you on the distortion figures....Some people have a love/hate thing with futurlec, but I've always gotten my stuff. Shipping takes a little bit though - I think they're in Singapore or something like that. ;)Gav. Quote
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.