Current divider mirror
in [Design]
|
From: JosephKK on 8 Feb 2010 00:38 On Sat, 6 Feb 2010 19:42:35 -0600, "Tim Williams" <tmoranwms(a)charter.net> wrote: >"JosephKK" <quiettechblue(a)yahoo.com> wrote in message >news:7anom5d66k5nn9k3epml4tm5q4hbuelh9g(a)4ax.com... >> Do you need continuously adjustable or can you live with discrete steps >> on the current ratio? > >Continuous, or sufficiently continuous (>= 8 bits?). > >I suppose an alternate formulation might be, if it can't be varied reliably >over a wide range, then it could be varied over a small dither range while >the switching handles the wide range, but that would quickly get ugly.. > >Tim For fully continuous i would look at a application of a Gilbert cell multiplier. There are some relatively exposed Gilbert cell ICs available. For a steppy solution an 8 or more bit (multiplying) current diverting DAC. Though i am not sure if a DAC will work well over two decades of reference current, you may have to subrange to two or three steps (that may overlap). Both solutions will require some appropriate scaling before or after or both.
From: JosephKK on 8 Feb 2010 00:43 On Sat, 06 Feb 2010 20:01:31 -0800, John Larkin <jjlarkin(a)highNOTlandTHIStechnologyPART.com> wrote: >On Sat, 06 Feb 2010 19:45:38 -0800, >"JosephKK"<quiettechblue(a)yahoo.com> wrote: > >>On Fri, 05 Feb 2010 21:02:46 -0800, John Larkin <jjlarkin(a)highNOTlandTHIStechnologyPART.com> wrote: >> >>>On Fri, 5 Feb 2010 11:21:05 -0600, "Tim Williams" >>><tmoranwms(a)charter.net> wrote: >>> >>>>"whit3rd" <whit3rd(a)gmail.com> wrote in message >>>>news:152e5f52-6af1-4cb3-aa39-0ecf74e30db1(a)q4g2000yqm.googlegroups.com.... >>>>> The schemes John Larkin gave are unipolar, and 'current from +V or -V' >>>>> sounds like bipolar to me. >>>> >>>>No, it's unipolar, I meant the currents are sourced from the +V rail, or >>>>sunk into the -V rail. >>>> >>>>Your circuit reminds me of, >>>>http://webpages.charter.net/dawill/tmoranwms/Circuits_2008/Current_Amplifer.gif >>>>which isn't to be left alone on the breadboard unless the transistors are on >>>>a fairly large heatsink. ;-) >>>> >>>>Tim >>> >>>Using the opamp rail currents gives essentially perfect crossover >>>behavior. At zero signal, both mirrors run at their Iq. As you apply >>>signal in one direction, one mirror current goes up and the other >>>stays at Iq. That's different from most class AB stages, where as one >>>side conducts, the other is turned off... which can lead to weird >>>transient distortions. >>> >>>John >> >>And to think i cribbed that trick from datasheets and application >>notes in 1965 to build my first self designed stereo. > >This config is ideal for driving current into inductive loads, like >MRI coils. The inherently high output impedance makes closed-loop >dynamics mostly independent of load. > >For audio, where you want a low output impedance, it's not ideal. The >fix is to apply massive negative feedback, which requires tons of GBW >to waste at higher frequencies. That was probably hard to do in 1965. > >Got a schematic? > >John I doubt if i still have the schematic. I should be able to do a fresh instance with current parts. In a couple of days as an LTspice file OK. The parts to the original may be unobtainium now.
From: Tim Williams on 8 Feb 2010 14:19 "JosephKK" <quiettechblue(a)yahoo.com> wrote in message news:8q7vm5hrekhi5rq86ffrbflv655tjg45nm(a)4ax.com... > For fully continuous i would look at a application of a Gilbert cell > multiplier. > There are some relatively exposed Gilbert cell ICs available. Hmm. I have some MC1496's laying around, that's halfway there. Or use a TIA. I wouldn't need four quadrants, one would be enough. That would even suggest a JFET. The problem is making the 0.1 to 10 ratio repeatable with an output from 50nA to 50mA (maybe not 50 on the highest range, I could settle for 5mA max.). That means putting the programming current into the tail is tricky (a regular current mirror would do okay), and the output can only be varied from 1 to 0.01 instead of 0.1 to 10 times (well, close enough). In this circuit, http://webpages.charter.net/dawill/tmoranwms/Circuits_2008/Triangle.gif I did something kinda similar, but more simply adjustable, and obviously the ratio is fixed. As I recall, duty cycle was fairly consistent across the range, and at cutoff it was on the order of nanoamps through the 2N440x's. If the multiplier's ratio is constant for a given differential voltage offset (which will have to be temperature compensated, but that would simply mean sourcing that offset voltage from a diode), then this might work fairly well. Thanks, this is something to think about. Tim -- Deep Friar: a very philosophical monk. Website: http://webpages.charter.net/dawill/tmoranwms
From: John Larkin on 8 Feb 2010 15:58 On Mon, 8 Feb 2010 13:19:43 -0600, "Tim Williams" <tmoranwms(a)charter.net> wrote: >"JosephKK" <quiettechblue(a)yahoo.com> wrote in message >news:8q7vm5hrekhi5rq86ffrbflv655tjg45nm(a)4ax.com... >> For fully continuous i would look at a application of a Gilbert cell >> multiplier. >> There are some relatively exposed Gilbert cell ICs available. > >Hmm. I have some MC1496's laying around, that's halfway there. Or use a >TIA. I wouldn't need four quadrants, one would be enough. That would even >suggest a JFET. The problem is making the 0.1 to 10 ratio repeatable with >an output from 50nA to 50mA (maybe not 50 on the highest range, I could >settle for 5mA max.). That means putting the programming current into the >tail is tricky (a regular current mirror would do okay), and the output can >only be varied from 1 to 0.01 instead of 0.1 to 10 times (well, close >enough). > >In this circuit, >http://webpages.charter.net/dawill/tmoranwms/Circuits_2008/Triangle.gif That circuit is so astoundingly bad that it's worth saving. John
From: Tim Williams on 8 Feb 2010 16:26
"John Larkin" <jjlarkin(a)highNOTlandTHIStechnologyPART.com> wrote in message news:3ju0n51fhfrj2pukuet0obtibm2p7lnmu3(a)4ax.com... >>http://webpages.charter.net/dawill/tmoranwms/Circuits_2008/Triangle.gif > > That circuit is so astoundingly bad that it's worth saving. Bad for what? It has an unusually wide range, as discrete circuits go. On 680pF, I got from 1Hz to 4.5MHz in a single range. Six decades isn't bad for a 2N440x. Smaller RF BJT's would go lower, or you could do range switching or more exotic approaches (leaky photodiode?). Besides the wide range, what impressed me is the diffamp got the 2N4403 switching in about 20ns, for a balls-out maximum frequency around 35MHz. Faster transistors would also take this arbitrarily high (the PHEMTs you're so fond of would probably make a proper RC oscillator in the GHz). Say, do they even make P type stupidfast transistors, SiGe or otherwise? I remember they don't bother with P type GaAs or InP since they suck for holes. I don't remember if, when set at ~10Hz or so, the frequency drifted by a decade or so when I touched the leftmost transistor. It ought to. I do remember seeing it change in steps, since I was using a wirewound pot. Tim -- Deep Friar: a very philosophical monk. Website: http://webpages.charter.net/dawill/tmoranwms |