I have a differential output from a fast CML flipflop, powered by +3 and ground, and want to drive another CML-input part. If the load gadget was also powered from +3, I'd just connect them with a couple of 50 ohm traces. But the Vcc of the destination part could be anything from +3 to -2. I want full speed and DC coupling, so the ideal part to put in series with both runs is a battery of the appropriate voltage, namely the difference in supply voltages. Couldn't find anything that would work like that. This thing below sorta fakes the batteries. It seems to work. The constraints on the CML transmitter (how far can the pins actually swing?) and on the receiver device (how far can the pins actually swing?) are far from clear, so we'll have to test this some. This looks pretty simple, but took a lot of thinking and many stupid simulations to get it to be simple. It bothered me enough to keep at it. Can't ski... too much snow. Version 4 SHEET 1 2664 1128 WIRE 48 32 -144 32 WIRE 128 32 48 32 WIRE 320 32 288 32 WIRE 352 32 320 32 WIRE 544 32 496 32 WIRE 576 32 544 32 WIRE 848 32 736 32 WIRE 944 32 848 32 WIRE 1328 32 1296 32 WIRE 1360 32 1328 32 WIRE 1664 32 1632 32 WIRE 1696 32 1664 32 WIRE -144 80 -144 32 WIRE 944 80 944 32 WIRE 1296 80 1296 32 WIRE 1632 80 1632 32 WIRE 128 144 128 32 WIRE 288 144 288 32 WIRE 496 144 496 32 WIRE 736 144 736 32 WIRE 1296 192 1296 160 WIRE 1632 192 1632 160 WIRE -144 224 -144 160 WIRE 944 224 944 160 WIRE -32 368 -144 368 WIRE 128 368 128 224 WIRE 128 368 48 368 WIRE 224 368 128 368 WIRE 288 368 288 224 WIRE 288 368 224 368 WIRE 352 368 288 368 WIRE 496 368 496 224 WIRE 496 368 416 368 WIRE 608 368 496 368 WIRE 736 368 736 224 WIRE 736 368 608 368 WIRE 1328 416 1296 416 WIRE 1360 416 1328 416 WIRE 1664 416 1632 416 WIRE 1696 416 1664 416 WIRE -144 448 -144 368 WIRE 1296 464 1296 416 WIRE 1632 464 1632 416 WIRE 1216 480 1200 480 WIRE 1264 480 1216 480 WIRE 1600 480 1488 480 WIRE 1376 496 1328 496 WIRE 1392 496 1376 496 WIRE 1712 496 1664 496 WIRE 1728 496 1712 496 WIRE 1264 512 1152 512 WIRE 1488 512 1488 480 WIRE 1552 512 1536 512 WIRE 1600 512 1552 512 WIRE 1152 544 1152 512 WIRE 1296 576 1296 528 WIRE 1328 576 1296 576 WIRE 1360 576 1328 576 WIRE 1632 576 1632 528 WIRE 1664 576 1632 576 WIRE 1696 576 1664 576 WIRE -32 768 -144 768 WIRE 128 768 48 768 WIRE 224 768 128 768 WIRE 304 768 224 768 WIRE 352 768 304 768 WIRE 496 768 416 768 WIRE 608 768 496 768 WIRE 736 768 608 768 WIRE 1344 816 1296 816 WIRE 1376 816 1344 816 WIRE 1712 816 1632 816 WIRE 1744 816 1712 816 WIRE -144 864 -144 768 WIRE 1296 864 1296 816 WIRE 1632 864 1632 816 WIRE 128 880 128 768 WIRE 304 880 304 768 WIRE 496 880 496 768 WIRE 736 880 736 768 WIRE 1216 880 1184 880 WIRE 1248 880 1216 880 WIRE 1552 880 1536 880 WIRE 1584 880 1552 880 WIRE 1216 928 1184 928 WIRE 1248 928 1216 928 WIRE 1552 928 1536 928 WIRE 1584 928 1552 928 WIRE 1296 976 1296 944 WIRE 1632 976 1632 944 WIRE 64 1056 16 1056 WIRE 128 1056 128 960 WIRE 128 1056 64 1056 WIRE 304 1056 304 960 WIRE 352 1056 304 1056 WIRE 384 1056 352 1056 WIRE 496 1056 496 960 WIRE 528 1056 496 1056 WIRE 560 1056 528 1056 WIRE 736 1056 736 960 WIRE 816 1056 736 1056 WIRE 880 1056 816 1056 FLAG -144 224 0 FLAG 944 224 0 FLAG 224 768 FF- FLAG 848 32 VH FLAG 48 32 Vcc FLAG 64 1056 Vcc FLAG -144 448 0 FLAG -144 864 0 FLAG 816 1056 VH FLAG 224 368 FF+ FLAG 608 768 SY- FLAG 608 368 SY+ FLAG 320 32 A FLAG 544 32 B FLAG 528 1056 A FLAG 352 1056 B FLAG 1632 976 0 FLAG 1712 816 DIFF FLAG 1552 880 SY+ FLAG 1552 928 SY- FLAG 1296 192 0 FLAG 1632 192 0 FLAG 1328 32 AH FLAG 1664 32 AL FLAG 1296 976 0 FLAG 1344 816 GO FLAG 1216 880 FF+ FLAG 1216 928 FF- FLAG 1328 416 AH FLAG 1328 576 AL FLAG 1664 416 AH FLAG 1664 576 AL FLAG 1376 496 A FLAG 1712 496 B FLAG 1152 544 0 FLAG 1216 480 GO FLAG 1488 512 0 FLAG 1552 512 GO SYMBOL res 112 128 R0 WINDOW 0 -64 41 Left 2 WINDOW 3 -63 73 Left 2 SYMATTR InstName R1 SYMATTR Value 50 SYMBOL current 48 368 R90 WINDOW 0 -97 29 VRight 2 WINDOW 3 -57 -54 VRight 2 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 SYMATTR InstName Icml SYMATTR Value PULSE(16m 0 1u 5n 1n 100u) SYMBOL voltage -144 64 R0 WINDOW 0 65 38 Left 2 WINDOW 3 72 68 Left 2 SYMATTR InstName V1 SYMATTR Value 3 SYMBOL res 720 128 R0 WINDOW 0 62 38 Left 2 WINDOW 3 63 74 Left 2 SYMATTR InstName R2 SYMATTR Value 50 SYMBOL voltage 944 64 R0 WINDOW 0 -86 34 Left 2 WINDOW 3 -77 61 Left 2 SYMATTR InstName VH SYMATTR Value 2 SYMBOL res 112 864 R0 WINDOW 0 -62 35 Left 2 WINDOW 3 -62 69 Left 2 SYMATTR InstName R3 SYMATTR Value 50 SYMBOL bi 48 768 R90 WINDOW 0 -91 40 VBottom 2 WINDOW 3 -77 44 VTop 2 SYMATTR InstName B1 SYMATTR Value I=16m-I(Icml) SYMBOL cap 416 352 R90 WINDOW 0 -43 31 VBottom 2 WINDOW 3 -35 32 VTop 2 SYMATTR InstName C1 SYMATTR Value 1� SYMBOL cap 416 752 R90 WINDOW 0 70 29 VBottom 2 WINDOW 3 77 27 VTop 2 SYMATTR InstName C3 SYMATTR Value 1� SYMBOL res 720 864 R0 WINDOW 0 62 42 Left 2 WINDOW 3 62 71 Left 2 SYMATTR InstName R9 SYMATTR Value 50 SYMBOL res 304 240 R180 WINDOW 0 -51 69 Left 2 WINDOW 3 -62 38 Left 2 SYMATTR InstName R4 SYMATTR Value 1.5K SYMBOL res 480 128 R0 WINDOW 0 56 40 Left 2 WINDOW 3 48 73 Left 2 SYMATTR InstName R5 SYMATTR Value 1.5K SYMBOL res 320 976 R180 WINDOW 0 -51 69 Left 2 WINDOW 3 -60 37 Left 2 SYMATTR InstName R7 SYMATTR Value 1.5K SYMBOL res 480 864 R0 WINDOW 0 58 35 Left 2 WINDOW 3 48 70 Left 2 SYMATTR InstName R8 SYMATTR Value 1.5K SYMBOL e 1632 848 R0 WINDOW 0 53 41 Left 2 WINDOW 3 60 67 Left 2 SYMATTR InstName E2 SYMATTR Value 1 SYMBOL Opamps\\UniversalOpamp2 1296 496 R0 SYMATTR InstName U2 SYMATTR Value2 Avol=1Meg GBW=10Meg Slew=20Meg SYMBOL voltage 1296 64 R0 WINDOW 0 52 41 Left 2 WINDOW 3 60 70 Left 2 SYMATTR InstName V2 SYMATTR Value 6 SYMBOL voltage 1632 64 R0 WINDOW 0 55 36 Left 2 WINDOW 3 53 67 Left 2 SYMATTR InstName V3 SYMATTR Value -6 SYMBOL e 1296 848 R0 WINDOW 0 51 33 Left 2 WINDOW 3 55 63 Left 2 SYMATTR InstName E1 SYMATTR Value 1 SYMBOL Opamps\\UniversalOpamp2 1632 496 R0 SYMATTR InstName U1 SYMATTR Value2 Avol=1Meg GBW=10Meg Slew=20Meg TEXT -64 560 Left 2 ;CML Flop TEXT 800 560 Left 2 ;CML Load TEXT 264 504 Left 3 ;CML-CML Level Shifter TEXT 328 616 Left 2 !.tran 0 200u 0 TEXT 320 560 Left 2 ;JL Dec 24 2021 TEXT 896 272 Left 2 ;+3 to -2 RECTANGLE Normal 176 1120 -224 -32 2 RECTANGLE Normal 1040 1120 640 -32 2 -- If a man will begin with certainties, he shall end with doubts, but if he will be content to begin with doubts he shall end in certainties. Francis Bacon
CML-CML level shifter
Started by ●December 24, 2021
Reply by ●December 25, 20212021-12-25
On Friday, December 24, 2021 at 2:21:29 PM UTC-8, John Larkin wrote:> I have a differential output from a fast CML flipflop, powered by +3 > and ground, and want to drive another CML-input part. If the load > gadget was also powered from +3, I'd just connect them with a couple > of 50 ohm traces. But the Vcc of the destination part could be > anything from +3 to -2. > > I want full speed and DC coupling, so the ideal part to put in series > with both runs is a battery of the appropriate voltage,So, do that. Just a capacitor, obviously, won't have the constant DC step if there's any low frequencies present, but you can put a three-transistor current mirror dual-source on the most positive rail, and another three-transistor current mirror dual-sink on the most negative rail, and connect the input transistors' bases of those mirrors with the appropriate resistor to make a constant-current bias. Then, instead of coupling capacitor alone, you put a bit of bypass resistor across each capacitor, and feed the high terminal with one source, and the low terminal with one sink (and 'cuz it's differential, you'd have the second source and sink for the other half's capacitor). As long as you don't dial the current up beyond what the CML sources , it'll drive a lot like a battery in series. If the HF gets too much Miller effect, a few ferrite beads can help, obviously. You do have to know polarity of the offset required at wiring-time, but the mirrors' emitter supplies are places you can apply dynamic controls of the amplitude and range of offset.
Reply by ●December 25, 20212021-12-25
On Fri, 24 Dec 2021 20:47:01 -0800 (PST), whit3rd <whit3rd@gmail.com> wrote:>On Friday, December 24, 2021 at 2:21:29 PM UTC-8, John Larkin wrote: >> I have a differential output from a fast CML flipflop, powered by +3 >> and ground, and want to drive another CML-input part. If the load >> gadget was also powered from +3, I'd just connect them with a couple >> of 50 ohm traces. But the Vcc of the destination part could be >> anything from +3 to -2. >> >> I want full speed and DC coupling, so the ideal part to put in series >> with both runs is a battery of the appropriate voltage, > >So, do that. Just a capacitor, obviously, won't have the constant DC step >if there's any low frequencies present, but you can put a three-transistor >current mirror dual-source on the most positive rail, and another three-transistor >current mirror dual-sink on the most negative rail, and connect the input >transistors' bases of those mirrors with the appropriate resistor to make >a constant-current bias. Then, instead of coupling capacitor alone, you >put a bit of bypass resistor across each capacitor, and feed the high terminal >with one source, and the low terminal with one sink (and 'cuz it's differential, you'd >have the second source and sink for the other half's capacitor). > >As long as you don't dial the current up beyond what the CML sources , it'll drive >a lot like a battery in series. If the HF gets too much Miller effect, a few ferrite beads >can help, obviously. > >You do have to know polarity of the offset required at wiring-time, > but the mirrors' emitter supplies are places you can apply >dynamic controls of the amplitude and range of offset.The polarity can go either way, which sure doesn't help. In this case, I want to couple logic levels with features from maybe 50 ps to 50 years. I might grudgingly allow the two traces between chips to be half an inch long, with one sideways cap in the middle of each. That's no place for a dozen transistors or ICs. My circuit started as two floating programmable voltages across the coupling caps, and stepwise deteriorated to the simple thing I have now. The big resistors fake dueling current sources, and the dropping resistor across the cap turns out to work at infinite ohms. There is a more general issue of splitting a signal into a number of bandwidths, transmitting, and then recombining neatly. Phil Hobbs recently needed to drive a ganfet gate with a fast-edge fairly long pulse, when the fet is riding hundreds of volts off ground. I don't know how he wound up doing that. In audio, a crossover has similar issues, bandwidth splitting. You can buy logic isolator chips, some with isolated power, but they are terrible for fast stuff. I've many times combined a fast (cap or transformer) AC path with a slow optocoupler. There are megabuck class oscilloscopes that split signals onto multiple paths of different bandwidths (with mixers I think) and then digitize and recombine with lots of DSP. That's the deconvolution problem. The general problem of bandwidth splitting and combining keeps popping up. It's worth a long paper or a short book. This is a good start: https://www.dropbox.com/s/ct7x6449fjz5d4q/RL-RC.jpg?raw=1 -- I yam what I yam - Popeye
Reply by ●December 25, 20212021-12-25
jlarkin@highlandsniptechnology.com wrote:> On Fri, 24 Dec 2021 20:47:01 -0800 (PST), whit3rd <whit3rd@gmail.com> > wrote: > >> On Friday, December 24, 2021 at 2:21:29 PM UTC-8, John Larkin wrote: >>> I have a differential output from a fast CML flipflop, powered by +3 >>> and ground, and want to drive another CML-input part. If the load >>> gadget was also powered from +3, I'd just connect them with a couple >>> of 50 ohm traces. But the Vcc of the destination part could be >>> anything from +3 to -2. >>> >>> I want full speed and DC coupling, so the ideal part to put in series >>> with both runs is a battery of the appropriate voltage, >> >> So, do that. Just a capacitor, obviously, won't have the constant DC step >> if there's any low frequencies present, but you can put a three-transistor >> current mirror dual-source on the most positive rail, and another three-transistor >> current mirror dual-sink on the most negative rail, and connect the input >> transistors' bases of those mirrors with the appropriate resistor to make >> a constant-current bias. Then, instead of coupling capacitor alone, you >> put a bit of bypass resistor across each capacitor, and feed the high terminal >> with one source, and the low terminal with one sink (and 'cuz it's differential, you'd >> have the second source and sink for the other half's capacitor). >> >> As long as you don't dial the current up beyond what the CML sources , it'll drive >> a lot like a battery in series. If the HF gets too much Miller effect, a few ferrite beads >> can help, obviously. >> >> You do have to know polarity of the offset required at wiring-time, >> but the mirrors' emitter supplies are places you can apply >> dynamic controls of the amplitude and range of offset. > > The polarity can go either way, which sure doesn't help. > > In this case, I want to couple logic levels with features from maybe > 50 ps to 50 years. I might grudgingly allow the two traces between > chips to be half an inch long, with one sideways cap in the middle of > each. That's no place for a dozen transistors or ICs. My circuit > started as two floating programmable voltages across the coupling > caps, and stepwise deteriorated to the simple thing I have now. The > big resistors fake dueling current sources, and the dropping resistor > across the cap turns out to work at infinite ohms. > > There is a more general issue of splitting a signal into a number of > bandwidths, transmitting, and then recombining neatly. Phil Hobbs > recently needed to drive a ganfet gate with a fast-edge fairly long > pulse, when the fet is riding hundreds of volts off ground. I don't > know how he wound up doing that.Hasn't been implemented yet. Probably something like this: <https://electrooptical.net/www/sed/IsolatedGaN.zip>. The Coilcraft transformer is way too clunky, so it'll either be an RC or a much lighter-weight transformer--maybe one of your coax-jumper-plus-potcore things. (The switch has to ride on top of a -450V avalanche photodiode bias supply.) The common-gate FET is an interesting idea I found in an app note, of all places. (I assume that means it's old hat, but I hadn't seen it before.) The substrate diode conducts in one direction and the transistor action in the other, so the hold time is limited by the RC and not the voltseconds of the transformer. It also allows extending the hold time indefinitely by pulsing the gate driver.> In audio, a crossover has similar > issues, bandwidth splitting. > > You can buy logic isolator chips, some with isolated power, but they > are terrible for fast stuff. I've many times combined a fast (cap or > transformer) AC path with a slow optocoupler. > > There are megabuck class oscilloscopes that split signals onto > multiple paths of different bandwidths (with mixers I think) and then > digitize and recombine with lots of DSP. That's the deconvolution > problem. > > The general problem of bandwidth splitting and combining keeps popping > up. It's worth a long paper or a short book. This is a good start: > > https://www.dropbox.com/s/ct7x6449fjz5d4q/RL-RC.jpg?raw=1I'd like to read that book! The Tektronix 'feedbeside' thing requires really good matching between branches to get better than oscilloscope accuracy. If the matching isn't very close, you wind up with low-frequency pole/zero pairs that don't quite cancel. That leads to settling whoopdedoos at late times, which are often super obnoxious in real applications but which will be quite invisible on a scope. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net http://hobbs-eo.com
Reply by ●December 25, 20212021-12-25
On Sat, 25 Dec 2021 17:57:39 -0500, Phil Hobbs <pcdhSpamMeSenseless@electrooptical.net> wrote:>jlarkin@highlandsniptechnology.com wrote: >> On Fri, 24 Dec 2021 20:47:01 -0800 (PST), whit3rd <whit3rd@gmail.com> >> wrote: >> >>> On Friday, December 24, 2021 at 2:21:29 PM UTC-8, John Larkin wrote: >>>> I have a differential output from a fast CML flipflop, powered by +3 >>>> and ground, and want to drive another CML-input part. If the load >>>> gadget was also powered from +3, I'd just connect them with a couple >>>> of 50 ohm traces. But the Vcc of the destination part could be >>>> anything from +3 to -2. >>>> >>>> I want full speed and DC coupling, so the ideal part to put in series >>>> with both runs is a battery of the appropriate voltage, >>> >>> So, do that. Just a capacitor, obviously, won't have the constant DC step >>> if there's any low frequencies present, but you can put a three-transistor >>> current mirror dual-source on the most positive rail, and another three-transistor >>> current mirror dual-sink on the most negative rail, and connect the input >>> transistors' bases of those mirrors with the appropriate resistor to make >>> a constant-current bias. Then, instead of coupling capacitor alone, you >>> put a bit of bypass resistor across each capacitor, and feed the high terminal >>> with one source, and the low terminal with one sink (and 'cuz it's differential, you'd >>> have the second source and sink for the other half's capacitor). >>> >>> As long as you don't dial the current up beyond what the CML sources , it'll drive >>> a lot like a battery in series. If the HF gets too much Miller effect, a few ferrite beads >>> can help, obviously. >>> >>> You do have to know polarity of the offset required at wiring-time, >>> but the mirrors' emitter supplies are places you can apply >>> dynamic controls of the amplitude and range of offset. >> >> The polarity can go either way, which sure doesn't help. >> >> In this case, I want to couple logic levels with features from maybe >> 50 ps to 50 years. I might grudgingly allow the two traces between >> chips to be half an inch long, with one sideways cap in the middle of >> each. That's no place for a dozen transistors or ICs. My circuit >> started as two floating programmable voltages across the coupling >> caps, and stepwise deteriorated to the simple thing I have now. The >> big resistors fake dueling current sources, and the dropping resistor >> across the cap turns out to work at infinite ohms. >> >> There is a more general issue of splitting a signal into a number of >> bandwidths, transmitting, and then recombining neatly. Phil Hobbs >> recently needed to drive a ganfet gate with a fast-edge fairly long >> pulse, when the fet is riding hundreds of volts off ground. I don't >> know how he wound up doing that. > >Hasn't been implemented yet. Probably something like this: ><https://electrooptical.net/www/sed/IsolatedGaN.zip>. > >The Coilcraft transformer is way too clunky, so it'll either be an RC or >a much lighter-weight transformer--maybe one of your >coax-jumper-plus-potcore things. (The switch has to ride on top of a >-450V avalanche photodiode bias supply.) > >The common-gate FET is an interesting idea I found in an app note, of >all places. (I assume that means it's old hat, but I hadn't seen it >before.) > >The substrate diode conducts in one direction and the transistor action >in the other, so the hold time is limited by the RC and not the >voltseconds of the transformer. It also allows extending the hold time >indefinitely by pulsing the gate driver. > >> In audio, a crossover has similar >> issues, bandwidth splitting. >> >> You can buy logic isolator chips, some with isolated power, but they >> are terrible for fast stuff. I've many times combined a fast (cap or >> transformer) AC path with a slow optocoupler. >> >> There are megabuck class oscilloscopes that split signals onto >> multiple paths of different bandwidths (with mixers I think) and then >> digitize and recombine with lots of DSP. That's the deconvolution >> problem. >> >> The general problem of bandwidth splitting and combining keeps popping >> up. It's worth a long paper or a short book. This is a good start: >> >> https://www.dropbox.com/s/ct7x6449fjz5d4q/RL-RC.jpg?raw=1 > >I'd like to read that book! > >The Tektronix 'feedbeside' thing requires really good matching between >branches to get better than oscilloscope accuracy. > >If the matching isn't very close, you wind up with low-frequency >pole/zero pairs that don't quite cancel. That leads to settling >whoopdedoos at late times, which are often super obnoxious in real >applications but which will be quite invisible on a scope.The equivalent of p/z mismatch is a problem even in logic couplers. The trick is to keep the corner freqs far apart. The AC path should have a much longer time constant than the maintaining DC path. A bit of clamping helps too.> >Cheers > >Phil HobbsWhen Tek did the 7104 1 GHz microchannel scope, they considered splitting the gain path into bandwidth zones, amplifying, and combining. But linear ICs got good enough about that time, so they used them. -- I yam what I yam - Popeye
Reply by ●December 25, 20212021-12-25
jlarkin@highlandsniptechnology.com wrote:> On Sat, 25 Dec 2021 17:57:39 -0500, Phil Hobbs > <pcdhSpamMeSenseless@electrooptical.net> wrote: > >> jlarkin@highlandsniptechnology.com wrote: >>> On Fri, 24 Dec 2021 20:47:01 -0800 (PST), whit3rd <whit3rd@gmail.com> >>> wrote: >>> >>>> On Friday, December 24, 2021 at 2:21:29 PM UTC-8, John Larkin wrote: >>>>> I have a differential output from a fast CML flipflop, powered by +3 >>>>> and ground, and want to drive another CML-input part. If the load >>>>> gadget was also powered from +3, I'd just connect them with a couple >>>>> of 50 ohm traces. But the Vcc of the destination part could be >>>>> anything from +3 to -2. >>>>> >>>>> I want full speed and DC coupling, so the ideal part to put in series >>>>> with both runs is a battery of the appropriate voltage, >>>> >>>> So, do that. Just a capacitor, obviously, won't have the constant DC step >>>> if there's any low frequencies present, but you can put a three-transistor >>>> current mirror dual-source on the most positive rail, and another three-transistor >>>> current mirror dual-sink on the most negative rail, and connect the input >>>> transistors' bases of those mirrors with the appropriate resistor to make >>>> a constant-current bias. Then, instead of coupling capacitor alone, you >>>> put a bit of bypass resistor across each capacitor, and feed the high terminal >>>> with one source, and the low terminal with one sink (and 'cuz it's differential, you'd >>>> have the second source and sink for the other half's capacitor). >>>> >>>> As long as you don't dial the current up beyond what the CML sources , it'll drive >>>> a lot like a battery in series. If the HF gets too much Miller effect, a few ferrite beads >>>> can help, obviously. >>>> >>>> You do have to know polarity of the offset required at wiring-time, >>>> but the mirrors' emitter supplies are places you can apply >>>> dynamic controls of the amplitude and range of offset. >>> >>> The polarity can go either way, which sure doesn't help. >>> >>> In this case, I want to couple logic levels with features from maybe >>> 50 ps to 50 years. I might grudgingly allow the two traces between >>> chips to be half an inch long, with one sideways cap in the middle of >>> each. That's no place for a dozen transistors or ICs. My circuit >>> started as two floating programmable voltages across the coupling >>> caps, and stepwise deteriorated to the simple thing I have now. The >>> big resistors fake dueling current sources, and the dropping resistor >>> across the cap turns out to work at infinite ohms. >>> >>> There is a more general issue of splitting a signal into a number of >>> bandwidths, transmitting, and then recombining neatly. Phil Hobbs >>> recently needed to drive a ganfet gate with a fast-edge fairly long >>> pulse, when the fet is riding hundreds of volts off ground. I don't >>> know how he wound up doing that. >> >> Hasn't been implemented yet. Probably something like this: >> <https://electrooptical.net/www/sed/IsolatedGaN.zip>. >> >> The Coilcraft transformer is way too clunky, so it'll either be an RC or >> a much lighter-weight transformer--maybe one of your >> coax-jumper-plus-potcore things. (The switch has to ride on top of a >> -450V avalanche photodiode bias supply.) >> >> The common-gate FET is an interesting idea I found in an app note, of >> all places. (I assume that means it's old hat, but I hadn't seen it >> before.) >> >> The substrate diode conducts in one direction and the transistor action >> in the other, so the hold time is limited by the RC and not the >> voltseconds of the transformer. It also allows extending the hold time >> indefinitely by pulsing the gate driver. >> >>> In audio, a crossover has similar >>> issues, bandwidth splitting. >>> >>> You can buy logic isolator chips, some with isolated power, but they >>> are terrible for fast stuff. I've many times combined a fast (cap or >>> transformer) AC path with a slow optocoupler. >>> >>> There are megabuck class oscilloscopes that split signals onto >>> multiple paths of different bandwidths (with mixers I think) and then >>> digitize and recombine with lots of DSP. That's the deconvolution >>> problem. >>> >>> The general problem of bandwidth splitting and combining keeps popping >>> up. It's worth a long paper or a short book. This is a good start: >>> >>> https://www.dropbox.com/s/ct7x6449fjz5d4q/RL-RC.jpg?raw=1 >> >> I'd like to read that book! >> >> The Tektronix 'feedbeside' thing requires really good matching between >> branches to get better than oscilloscope accuracy. >> >> If the matching isn't very close, you wind up with low-frequency >> pole/zero pairs that don't quite cancel. That leads to settling >> whoopdedoos at late times, which are often super obnoxious in real >> applications but which will be quite invisible on a scope. > > The equivalent of p/z mismatch is a problem even in logic couplers. > The trick is to keep the corner freqs far apart. The AC path should > have a much longer time constant than the maintaining DC path. A bit > of clamping helps too.I'm not sure that works in general, though--you'd need one feedback loop to govern the other, and you always wind up with problems where they change roles. Making the AC time constant slow just exports the problem to late times. Some applications don't care, in which case that's a big win, but some do. Cheers Phil Hobbs PS: Merry Christmas! -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net http://hobbs-eo.com
Reply by ●December 26, 20212021-12-26
On Sat, 25 Dec 2021 19:46:38 -0500, Phil Hobbs <pcdhSpamMeSenseless@electrooptical.net> wrote:>jlarkin@highlandsniptechnology.com wrote: >> On Sat, 25 Dec 2021 17:57:39 -0500, Phil Hobbs >> <pcdhSpamMeSenseless@electrooptical.net> wrote: >> >>> jlarkin@highlandsniptechnology.com wrote: >>>> On Fri, 24 Dec 2021 20:47:01 -0800 (PST), whit3rd <whit3rd@gmail.com> >>>> wrote: >>>> >>>>> On Friday, December 24, 2021 at 2:21:29 PM UTC-8, John Larkin wrote: >>>>>> I have a differential output from a fast CML flipflop, powered by +3 >>>>>> and ground, and want to drive another CML-input part. If the load >>>>>> gadget was also powered from +3, I'd just connect them with a couple >>>>>> of 50 ohm traces. But the Vcc of the destination part could be >>>>>> anything from +3 to -2. >>>>>> >>>>>> I want full speed and DC coupling, so the ideal part to put in series >>>>>> with both runs is a battery of the appropriate voltage, >>>>> >>>>> So, do that. Just a capacitor, obviously, won't have the constant DC step >>>>> if there's any low frequencies present, but you can put a three-transistor >>>>> current mirror dual-source on the most positive rail, and another three-transistor >>>>> current mirror dual-sink on the most negative rail, and connect the input >>>>> transistors' bases of those mirrors with the appropriate resistor to make >>>>> a constant-current bias. Then, instead of coupling capacitor alone, you >>>>> put a bit of bypass resistor across each capacitor, and feed the high terminal >>>>> with one source, and the low terminal with one sink (and 'cuz it's differential, you'd >>>>> have the second source and sink for the other half's capacitor). >>>>> >>>>> As long as you don't dial the current up beyond what the CML sources , it'll drive >>>>> a lot like a battery in series. If the HF gets too much Miller effect, a few ferrite beads >>>>> can help, obviously. >>>>> >>>>> You do have to know polarity of the offset required at wiring-time, >>>>> but the mirrors' emitter supplies are places you can apply >>>>> dynamic controls of the amplitude and range of offset. >>>> >>>> The polarity can go either way, which sure doesn't help. >>>> >>>> In this case, I want to couple logic levels with features from maybe >>>> 50 ps to 50 years. I might grudgingly allow the two traces between >>>> chips to be half an inch long, with one sideways cap in the middle of >>>> each. That's no place for a dozen transistors or ICs. My circuit >>>> started as two floating programmable voltages across the coupling >>>> caps, and stepwise deteriorated to the simple thing I have now. The >>>> big resistors fake dueling current sources, and the dropping resistor >>>> across the cap turns out to work at infinite ohms. >>>> >>>> There is a more general issue of splitting a signal into a number of >>>> bandwidths, transmitting, and then recombining neatly. Phil Hobbs >>>> recently needed to drive a ganfet gate with a fast-edge fairly long >>>> pulse, when the fet is riding hundreds of volts off ground. I don't >>>> know how he wound up doing that. >>> >>> Hasn't been implemented yet. Probably something like this: >>> <https://electrooptical.net/www/sed/IsolatedGaN.zip>. >>> >>> The Coilcraft transformer is way too clunky, so it'll either be an RC or >>> a much lighter-weight transformer--maybe one of your >>> coax-jumper-plus-potcore things. (The switch has to ride on top of a >>> -450V avalanche photodiode bias supply.) >>> >>> The common-gate FET is an interesting idea I found in an app note, of >>> all places. (I assume that means it's old hat, but I hadn't seen it >>> before.) >>> >>> The substrate diode conducts in one direction and the transistor action >>> in the other, so the hold time is limited by the RC and not the >>> voltseconds of the transformer. It also allows extending the hold time >>> indefinitely by pulsing the gate driver. >>> >>>> In audio, a crossover has similar >>>> issues, bandwidth splitting. >>>> >>>> You can buy logic isolator chips, some with isolated power, but they >>>> are terrible for fast stuff. I've many times combined a fast (cap or >>>> transformer) AC path with a slow optocoupler. >>>> >>>> There are megabuck class oscilloscopes that split signals onto >>>> multiple paths of different bandwidths (with mixers I think) and then >>>> digitize and recombine with lots of DSP. That's the deconvolution >>>> problem. >>>> >>>> The general problem of bandwidth splitting and combining keeps popping >>>> up. It's worth a long paper or a short book. This is a good start: >>>> >>>> https://www.dropbox.com/s/ct7x6449fjz5d4q/RL-RC.jpg?raw=1 >>> >>> I'd like to read that book! >>> >>> The Tektronix 'feedbeside' thing requires really good matching between >>> branches to get better than oscilloscope accuracy. >>> >>> If the matching isn't very close, you wind up with low-frequency >>> pole/zero pairs that don't quite cancel. That leads to settling >>> whoopdedoos at late times, which are often super obnoxious in real >>> applications but which will be quite invisible on a scope. >> >> The equivalent of p/z mismatch is a problem even in logic couplers. >> The trick is to keep the corner freqs far apart. The AC path should >> have a much longer time constant than the maintaining DC path. A bit >> of clamping helps too. > >I'm not sure that works in general, though--you'd need one feedback loop >to govern the other, and you always wind up with problems where they >change roles. Making the AC time constant slow just exports the problem >to late times. Some applications don't care, in which case that's a big >win, but some do. > >Cheers > >Phil Hobbs > >PS: Merry Christmas!I got a beautiful set of razor sharp Wusthof knives. They slice through Tartine sourdough like a chain saw. Pay no mind to the band-aids. I got Mo a 48" Sony oled TV. It's astonishing. It makes the old LCD look washed-out and fuzzy. -- I yam what I yam - Popeye
Reply by ●December 26, 20212021-12-26
On Saturday, December 25, 2021 at 7:12:18 AM UTC-8, jla...@highlandsniptechnology.com wrote:> On Fri, 24 Dec 2021 20:47:01 -0800 (PST), whit3rd <whi...@gmail.com> > wrote: > >On Friday, December 24, 2021 at 2:21:29 PM UTC-8, John Larkin wrote: > >> I have a differential output from a fast CML flipflop, powered by +3 > >> and ground, and want to drive another CML-input part. If the load > >> gadget was also powered from +3, I'd just connect them with a couple > >> of 50 ohm traces. But the Vcc of the destination part could be > >> anything from +3 to -2. > >> > >> I want full speed and DC coupling, so the ideal part to put in series > >> with both runs is a battery of the appropriate voltage, > > > >So, do that. Just a capacitor, obviously, won't have the constant DC step > >if there's any low frequencies present, but you can put a three-transistor > >current mirror dual-source on the most positive rail, and...> The polarity can go either way, which sure doesn't help. > > In this case, I want to couple logic levels with features from maybe > 50 ps to 50 years. I might grudgingly allow the two traces between > chips to be half an inch long, with one sideways cap in the middle of > each. That's no place for a dozen transistors or ICs.It only has one resistor and capacitor in each signal wire, that other circuitry is just the power supply, carries (ideally) none of the HF signal; ferrites for blocking are... tiny, if required. The resistor could piggyback on the capacitor, if space is that tight...> The general problem of bandwidth splitting and combining keeps popping > up. It's worth a long paper or a short book.Oh, there's a lot of interesting problems; howzabout an injectable circular transmission line, so you can insert a transient signal and keep recirculating the signal past a refresh amplifier/sampler until those golden nanoseconds have had all the picoseconds nailed down in memory? You can't quite do that with a cyclotron, the electrons DO revisit regularly, but injection is always in tiny bunches... Some of the inline amplifiers for optical fiber are suggestive, though.
Reply by ●December 26, 20212021-12-26
On Sun, 26 Dec 2021 02:34:39 -0800 (PST), whit3rd <whit3rd@gmail.com> wrote:>On Saturday, December 25, 2021 at 7:12:18 AM UTC-8, jla...@highlandsniptechnology.com wrote: >> On Fri, 24 Dec 2021 20:47:01 -0800 (PST), whit3rd <whi...@gmail.com> >> wrote: >> >On Friday, December 24, 2021 at 2:21:29 PM UTC-8, John Larkin wrote: >> >> I have a differential output from a fast CML flipflop, powered by +3 >> >> and ground, and want to drive another CML-input part. If the load >> >> gadget was also powered from +3, I'd just connect them with a couple >> >> of 50 ohm traces. But the Vcc of the destination part could be >> >> anything from +3 to -2. >> >> >> >> I want full speed and DC coupling, so the ideal part to put in series >> >> with both runs is a battery of the appropriate voltage, >> > >> >So, do that. Just a capacitor, obviously, won't have the constant DC step >> >if there's any low frequencies present, but you can put a three-transistor >> >current mirror dual-source on the most positive rail, and... > >> The polarity can go either way, which sure doesn't help. >> >> In this case, I want to couple logic levels with features from maybe >> 50 ps to 50 years. I might grudgingly allow the two traces between >> chips to be half an inch long, with one sideways cap in the middle of >> each. That's no place for a dozen transistors or ICs. > >It only has one resistor and capacitor in each signal wire, >that other circuitry is just the power supply, carries (ideally) >none of the HF signal; ferrites for blocking are... tiny, if required. > >The resistor could piggyback on the capacitor, if space is that tight...Got a schematic or a sim?> >> The general problem of bandwidth splitting and combining keeps popping >> up. It's worth a long paper or a short book. > >Oh, there's a lot of interesting problems; howzabout an injectable >circular transmission line, so you can insert a transient signal and keep recirculating >the signal past a refresh amplifier/sampler until those golden nanoseconds >have had all the picoseconds nailed down in memory? You can't quite >do that with a cyclotron, the electrons DO revisit regularly, but >injection is always in tiny bunches...We have considered a circular parametric oscillator as a breed of triggered oscillator, which would be cute but complex.> >Some of the inline amplifiers for optical fiber are suggestive, though.-- I yam what I yam - Popeye
Reply by ●December 26, 20212021-12-26
On Sat, 25 Dec 2021 17:57:39 -0500, Phil Hobbs <pcdhSpamMeSenseless@electrooptical.net> wrote:>jlarkin@highlandsniptechnology.com wrote: >> On Fri, 24 Dec 2021 20:47:01 -0800 (PST), whit3rd <whit3rd@gmail.com> >> wrote: >> >>> On Friday, December 24, 2021 at 2:21:29 PM UTC-8, John Larkin wrote: >>>> I have a differential output from a fast CML flipflop, powered by +3 >>>> and ground, and want to drive another CML-input part. If the load >>>> gadget was also powered from +3, I'd just connect them with a couple >>>> of 50 ohm traces. But the Vcc of the destination part could be >>>> anything from +3 to -2. >>>> >>>> I want full speed and DC coupling, so the ideal part to put in series >>>> with both runs is a battery of the appropriate voltage, >>> >>> So, do that. Just a capacitor, obviously, won't have the constant DC step >>> if there's any low frequencies present, but you can put a three-transistor >>> current mirror dual-source on the most positive rail, and another three-transistor >>> current mirror dual-sink on the most negative rail, and connect the input >>> transistors' bases of those mirrors with the appropriate resistor to make >>> a constant-current bias. Then, instead of coupling capacitor alone, you >>> put a bit of bypass resistor across each capacitor, and feed the high terminal >>> with one source, and the low terminal with one sink (and 'cuz it's differential, you'd >>> have the second source and sink for the other half's capacitor). >>> >>> As long as you don't dial the current up beyond what the CML sources , it'll drive >>> a lot like a battery in series. If the HF gets too much Miller effect, a few ferrite beads >>> can help, obviously. >>> >>> You do have to know polarity of the offset required at wiring-time, >>> but the mirrors' emitter supplies are places you can apply >>> dynamic controls of the amplitude and range of offset. >> >> The polarity can go either way, which sure doesn't help. >> >> In this case, I want to couple logic levels with features from maybe >> 50 ps to 50 years. I might grudgingly allow the two traces between >> chips to be half an inch long, with one sideways cap in the middle of >> each. That's no place for a dozen transistors or ICs. My circuit >> started as two floating programmable voltages across the coupling >> caps, and stepwise deteriorated to the simple thing I have now. The >> big resistors fake dueling current sources, and the dropping resistor >> across the cap turns out to work at infinite ohms. >> >> There is a more general issue of splitting a signal into a number of >> bandwidths, transmitting, and then recombining neatly. Phil Hobbs >> recently needed to drive a ganfet gate with a fast-edge fairly long >> pulse, when the fet is riding hundreds of volts off ground. I don't >> know how he wound up doing that. > >Hasn't been implemented yet. Probably something like this: ><https://electrooptical.net/www/sed/IsolatedGaN.zip>. > >The Coilcraft transformer is way too clunky, so it'll either be an RC or >a much lighter-weight transformer--maybe one of your >coax-jumper-plus-potcore things. (The switch has to ride on top of a >-450V avalanche photodiode bias supply.)You could maybe use a memory element on the high side, a flipflop or a schmitt trigger or maybe even just a capacitor, and drive it from narrow positive and negative spikes through a transformer. That could be done with zero static power required on the high side. The non-inverting schmitt case has analogies to my pulse coupler. Cap to the schmitt input, feedback resistor to sustain. Might work. -- I yam what I yam - Popeye