An Attempt to Quantify Semiconductor Differences

Discuss x0x construction and related issues

Moderators: altitude, adafruit_support_bill, adafruit, phono, hamburgers

Please be positive and constructive with your questions and comments.
Locked
User avatar
antto
 
Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: An Attempt to Quantify Semiconductor Differences

Post by antto »

i'm guessing this JFET changes the sawtooth amplitude and offset?

Brassteacher
 
Posts: 248
Joined: Fri Oct 08, 2010 4:49 am

Re: An Attempt to Quantify Semiconductor Differences

Post by Brassteacher »

antto wrote:i'm guessing this JFET changes the sawtooth amplitude and offset?
No and yes, sorta. The DC offset went from +7.98V with the J201 in place to +9.34V with the K30A-GR in place, quite a jump in DC offset.

As for the amplitude, that really can't change due to the sawtooth being pulled from the source of the JFET. In other words, the JFET functions as a non-inverting buffer in this case, and since it is pulled off the source, there is no voltage gain. In fact, no matter what you put there, bipolar, JFET, MOSFET, or even a triode (tube) the voltage gain will always be less than 1 (unity gain).

However, in this case the amplitude of the sawtooth DROPS with the K30A-GR in place as opposed to the J201. The reason being that the higher offset pushes the sawtooth too close to the positive rail, past the point of clipping. Luckily, it's not so far up that it clips hard. Pics below:

Oscilloscope probe AC coupled, 200Hz. J201 on left, K30A-GR on the right:
200Hz Saw JFET AC Coupled.jpg
200Hz Saw JFET AC Coupled.jpg (55.72 KiB) Viewed 4621 times
Probe DC coupled, 200Hz, J201 on left, K30A-GR on the right. Trace adjusted to center the J201 sawtooth on the X-axis:
200Hz Saw JFET DC Coupled.jpg
200Hz Saw JFET DC Coupled.jpg (54.94 KiB) Viewed 4621 times
In all cases above, the V/div. setting was not changed during the test, so the percentage of amplitude change or change in DC offset is valid. DC offset was also measured with a DMM.

Bottom line opinion: In case anyone reading needed it, this gives you even MORE reason to not use a JFET with a high Idss! For those that may not be familiar with the term "Idss", it's short for "current from drain to source, saturated". In other words, since I measured these JFETs at 9 Volts, their Idss rating is simply the maximum amount of current that can flow through the JFET at 9 Volts, when it is turned FULLY on. The Idss of a JFET, ideally, won't change much if at all with voltage when it is turned fully on, unless the voltage gets so low that the JFET stops conducting, or the voltage gets so high that the JFET blows up. For this reason, JFETs are great for using as a constant-current source, as in the case of Q39 in the x0x/TB-303. If you tie the gate to the source directly, the JFET turns fully on.

User avatar
antto
 
Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: An Attempt to Quantify Semiconductor Differences

Post by antto »

if anything changes the sawtooth offset and amplitude (or shape) it reacts on how the square wave would end up

one of my observations:
on most x0x'es, the amplitude of the squarewave compared to the sawtooth, is lower compared to that on a TB-303
also, the pulsewidth behaviour is slightly different
i mean, the x0x square has the same characteristics but slightly off

Brassteacher
 
Posts: 248
Joined: Fri Oct 08, 2010 4:49 am

Re: An Attempt to Quantify Semiconductor Differences

Post by Brassteacher »

antto wrote:if anything changes the sawtooth offset and amplitude (or shape) it reacts on how the square wave would end up

one of my observations:
on most x0x'es, the amplitude of the squarewave compared to the sawtooth, is lower compared to that on a TB-303
also, the pulsewidth behaviour is slightly different
i mean, the x0x square has the same characteristics but slightly off
Interesting. How does it differ? Perhaps one of the things that can come out of all this is finding out what to adjust to make the x0x more like the TB-303. Even from just a few quick 'scope screenshots and a few Idss measurements, we have a pretty good idea what changes the pulse-width and amplitude. I have a feeling that if a way to adjust the DC offset of the VCO were added, there would be much less change in the sawtooth characteristics no matter what JFET is installed, while still being able to "tweak" the squarewave.

I am beginning to think all this would be easier if I also had access to a refurbished TB-303. By refurbished I mean replacing all the electrolytic caps with new ones with like or similar specs to the originals, and checking for any other components that may have drifted significantly off value over time.

User avatar
antto
 
Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: An Attempt to Quantify Semiconductor Differences

Post by antto »

i think the 303 square pulsewidth is more "lose" in the lows
in other words: it gets thinner
another very important factor (soundwise) is the PW crosspoint frequency
at low frequencies, the PW is one way, then, at some point it is 50:50, and then above that frequency, the PW goes in the other direction
Image :roll:

User avatar
antto
 
Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: An Attempt to Quantify Semiconductor Differences

Post by antto »

the thing is, this waveshaper circuit, it has too many RC elements
so you get 303s with all kinds of small variations of the square wave
but the x0xb0xes usually all are a bit off, they all sound almost identical to one another, but not quite like the 303s in that matter
so i guess it's either some transistor, or the supply voltage (C11 charges from 12v and then there is 5.33 at the bottom (Q8 output)) or the tolerance of the RC elements
or a little bit from all

btw, you get a slightly better idea what i'm talking about if you somehow tweak the value of C11 (i can in my digital "model")
this is the first thing that impacts the PW
but the sawtooth offset and amplitude add to that
the 12v supply on C11 "scales" it in a way too
i'm not sure exactly how does the 5.33v supply changes the picture (as my "model" is not a real simulation)
but it reminds me of: the square wave on the x0xb0x compared to the sawtooth is quieter, compared to that on the 303

User avatar
altitude
 
Posts: 995
Joined: Wed May 11, 2005 5:17 pm

Re: An Attempt to Quantify Semiconductor Differences

Post by altitude »

antto wrote:...
but the x0xb0xes usually all are a bit off, they all sound almost identical to one another ...
I dont agree with that "sounding identical to one another" bit at all.

jayceefr
 
Posts: 21
Joined: Mon Mar 31, 2008 7:40 pm

Re: An Attempt to Quantify Semiconductor Differences

Post by jayceefr »

C11 seems a good idea. I just tried and replaced it with an old used tantalum I had lying around, and i prefer it nouw a lot. I find it sounds more round.

User avatar
antto
 
Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: An Attempt to Quantify Semiconductor Differences

Post by antto »

BANNED wrote:C11 seems a good idea. I just tried and replaced it with an old used tantalum I had lying around, and i prefer it nouw a lot. I find it sounds more round.
"round"
Do post before/after comparison, otherwise it's not much useful

Brassteacher
 
Posts: 248
Joined: Fri Oct 08, 2010 4:49 am

Re: An Attempt to Quantify Semiconductor Differences

Post by Brassteacher »

antto wrote:the thing is, this waveshaper circuit, it has too many RC elements
so you get 303s with all kinds of small variations of the square wave
but the x0xb0xes usually all are a bit off, they all sound almost identical to one another, but not quite like the 303s in that matter
so i guess it's either some transistor, or the supply voltage (C11 charges from 12v and then there is 5.33 at the bottom (Q8 output)) or the tolerance of the RC elements
or a little bit from all.
antto, I've been thinking about this, and studying the schematic. Besides variations in the 12V supply, a variation of Hfe in Q8 will affect waveshape and amplitude (I think we all know that), and variation in R36 will affect amplitude somewhat, and definitely DC offset of the square wave. This brings me to R22, which should affect gain (amplitude) when varied, and it appears R34 and R35 will affect pulse width and waveshape, as well as amplitude (possibly).

My question is, before I start installing trimpots, is it possible in your model to play with the values of R22, R34, and R35 to see which resistor affects which portion of the square wave parameters we're looking at? All three should have a significant amount of control over the charge/discharge time of C11, unless I had too much wine with dinner...

Also, thanks for putting the square wave drawings up. I think we may have been "counting" pulse width on the square wave differently, I look at the "upper" part of the wave, while I think you may have been looking at the "bottom" of the wave. In other words, we've been seeing the same thing, but how it is described changes depending if you are looking at the change in the bottom or top of the wave, which change opposite of each other.

Also, going back to the original TB-303 service manual, it appears that the sawtooth wave does in fact "bump" up against the 12V rail just a bit, knocking its amplitude down to 6.5V p-p. After finishing these recordings, I'll go back and adjust the 12V rail, as well as find a K30A-O out of my stash that most closely pushes the VCO output waveforms to match the ones shown in the service manual.

User avatar
antto
 
Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: An Attempt to Quantify Semiconductor Differences

Post by antto »

Brassteacher wrote:
antto wrote:the thing is, this waveshaper circuit, it has too many RC elements
so you get 303s with all kinds of small variations of the square wave
but the x0xb0xes usually all are a bit off, they all sound almost identical to one another, but not quite like the 303s in that matter
so i guess it's either some transistor, or the supply voltage (C11 charges from 12v and then there is 5.33 at the bottom (Q8 output)) or the tolerance of the RC elements
or a little bit from all.
antto, I've been thinking about this, and studying the schematic. Besides variations in the 12V supply, a variation of Hfe in Q8 will affect waveshape and amplitude (I think we all know that), and variation in R36 will affect amplitude somewhat, and definitely DC offset of the square wave. This brings me to R22, which should affect gain (amplitude) when varied, and it appears R34 and R35 will affect pulse width and waveshape, as well as amplitude (possibly).

My question is, before I start installing trimpots, is it possible in your model to play with the values of R22, R34, and R35 to see which resistor affects which portion of the square wave parameters we're looking at? All three should have a significant amount of control over the charge/discharge time of C11, unless I had too much wine with dinner...
my "model" is just an approximitation (not accurate really) of the transistor and the surrounding elements
probably the only accurate thing is the C11 LPF
my model got together by asking questions here: http://forums.adafruit.com/viewtopic.php?f=12&t=17355
more specifically from this post onwards: http://forums.adafruit.com/viewtopic.php?p=89078#p89078
Image
this was a LTspice simulation plotted at 2 different frequencies, but i set the plot width to match the periods of the waveform so it's easier to compare and see which parts are frequency dependant
the signal colors:
GREEN: the squarewave output (after Q8 collector)
BLUE: the naked sawtooth before it enters the waveshaper circuit
RED: sawtooth at the Q8 base (filtered by C10 R34 R45)
CYAN: 12V thru Q8 emitter

so what i did, i ran simulations at a lot of frequencies (from 5Hz up to 500Hz or something) and exported them into tables
then i used mIRC scripts to parse the exported tables and plot them on screen
and from there i started making a model, and ploted it against the simulation
at first i had absolutely no idea how a transistor works, how voltage and current works, i'm fairly dumb ;]
so i asked quest to explain, i also had a lot of help from the guys at #musicdsp
and i made a little algorithm which kinda mimics the transistor to some point
it has 3 different "modes" and it switches between them
i kinda started to get some similar results, but the model simply failed at a bunch of specific points
this means it wasn't the right algorithm
and i couldn't do better, so i "cheated" (used additional algos to compensate in the weak spots)
at the end, it's not really good
but it's better than what i had previously
the problem is: i don't understand how it works exactly, i have to be able to picture it in my head in order to come up with the algorithm
current, voltage, and transistors are still beyond me

the weakspots in my model are:
1) when Q8 goes into Fwd. Active mode - it doesn't instantly turn ON, it goes ON slowly, looks like a sigmoid shape, my model simply switches ON instantly, so i had to cheat here somehow (used an aditional LPF, but then if i set the freq. of this LPF to something which looks okay, it doesn't look okay at another plot of the simulation (at a different frequency) this means whatever causes this smooth transition from Cutoff to Fwd. Active is NOT a filter)
2) the Fwd. Active stage of the waveform, it's some form of an inverted version of the base signal with some offset, but again i had the same problem as above, when i tweak my model to look the same as the simulation plot, then i plot the simulation at another frequency - my model goes in a wrong direction, so i cheated again here, measured the error and compensated it additionally with an approximitation curve
3) the Saturation stage: this was the biggest difference
in the simulation, you can see how the RED and CYAN signals are "bent"
i think this is because of feedback going in all directions in the transistor
in my model iirc i got only 1 feedback loop, otherwise it gets too complex
i cheated again, i didn't ever fix the base signal in my model, it doesn't bent like this, it's only a filtered version of the BLUE signal
i did cheat in the CYAN signal tho, when Saturation begins, i start to feed some specific value to C11 in order to get it to look like the curve from the simulation

so, it's not at all accurate, but it does have some *similar* behaviour
and the resulting square wave changes alot when i change the parameters (C11 frequency, C10 frequency, 12V supply, Sawtooth offset and amplitude...)
Also, thanks for putting the square wave drawings up. I think we may have been "counting" pulse width on the square wave differently, I look at the "upper" part of the wave, while I think you may have been looking at the "bottom" of the wave. In other words, we've been seeing the same thing, but how it is described changes depending if you are looking at the change in the bottom or top of the wave, which change opposite of each other.
i'm not sure i understand you here
my point is, the pulse width changes with frequency
and at low freqs it's (for example) 20:80
then at some mid frequency it's 50:50
and then at higher freqs it goes in the other direction like 70:30
..and this is one of the very very audiable characters of the square wave in the 303
1) how much it goes away from 50:50 in the low freqs
2) where exactly does it cross 50:50 (at what frequency)

User avatar
antto
 
Posts: 1636
Joined: Thu Apr 15, 2010 3:21 pm

Re: An Attempt to Quantify Semiconductor Differences

Post by antto »

btw, guest posted the squarewave as a falstad circuit
you can import it there and look how it works, tweak the elements to see what happens
www.falstad.com/circuit/
when the applet opens, go to File Import and copy/paste the text there
guest wrote:here is my falstad export
it seems pretty accurate

$ 1 5.0E-6 10.20027730826997 78 5.0 50
t 320 224 288 224 0 -1 3.9060717001723395 0.5672672547949205 300.0
r 288 208 288 160 0 22000.0
r 288 240 288 304 0 10000.0
r 320 224 368 224 0 100000.0
r 320 224 320 176 0 10000.0
c 320 176 368 176 0 1.0E-8 0.12171641550847845
c 288 208 240 208 0 1.0E-6 8.671804446560614
g 240 208 240 224 0
R 288 160 288 128 0 0 40.0 11.8 0.0 0.0 0.5
R 368 176 416 176 0 4 130.0 -2.6 8.5 0.0 0.5
w 368 224 368 176 0
R 288 304 288 336 0 0 40.0 5.333 0.0 0.0 0.5
O 288 240 256 240 0
O 368 224 416 224 0
o 12 16 0 34 16.77994891013191 0.0 0 -1
o 13 128 0 34 20.0 9.765625E-5 1 -1

User avatar
aminoacid
 
Posts: 352
Joined: Tue Jun 20, 2006 5:27 am

Re: An Attempt to Quantify Semiconductor Differences

Post by aminoacid »

so first of all...

what happend with this thread?!


i really miss it, anto and brass! i wanna hear where you guys are in your experiments!!!

my self, seem to never close the box for good, even recently when i was quite saticfied i was going to close it after a good cleaning of the pcb. just by doing that gave me a totaly different character of the filter but also a faulty enevelope (zapped 4066 i think). of course when looking at my hef4066, what variations there was and what the original hd14066 looks like on schematics, i just had to order two differnt to see if thatll give some changes...

the xox projekt is a curse!


another thing that just came up to my interest is the thing with the diodes.

and just so you know, i dont know much about circuits and calculations at all...

look at d25 in the vco. im guessing it serves as a voltage drop in a bias circuit..?!
would this voltage drop/bias together with the voltage drop over q27 have anything to do with the waveshape (dc offset, transistors turning on/off timing, etc)

why im asking is cause my xox has
2sk30ao (tried 3 differnt and a 2sk118 without hearing much difference)
2sa733ap
1ss133 diodes

while my ml303 has got
j201
2sa733p
1n4148 diodes

-and the ml sounds actually sweeter in the vco.


knowing about the 1ss133 having less drops, i measured (diode test with dmm) the diodes in my xox and the ml303 and also others lying around. and it seems that theres a good swing in tolerance in the 4148. the ones in my ml has got smaller drops than the 1ss133 in my xox.

and last perception. is it by coincidence that d25 is called d25 and sits right next to q25 in the tb303? i dont know how they are mounted in the 303 but arent diodes some times placed tight together with transistors for temperature/Vbe/Ic stability. maybe the designer had this in mind atleast.

im going to my home town next week, and ill be bringing with me all my 1ss133 and make some experimantation but for now maybe you guys can say something about this...

but please dont shoot me cause im an amature :)

User avatar
aminoacid
 
Posts: 352
Joined: Tue Jun 20, 2006 5:27 am

Re: An Attempt to Quantify Semiconductor Differences

Post by aminoacid »

the diode thing didnt show anything of interest... pardon for making a fuzz about nothing...

however, q27 is much of interest. it seem to have some to say about the the waveshape/pulse width. i seem to like it really low beta as most other transistors in the xox... sawtooth sounds less sharp, more smooth with low beta. could be another variabel of the transistor also.

Locked
Please be positive and constructive with your questions and comments.

Return to “Making x0x”