Forums

Sampler diodes with more barrier height?

Started by Joerg February 15, 2014
On 02/17/2014 01:22 PM, John Larkin wrote:
> On Mon, 17 Feb 2014 08:34:44 -0800 (PST), bloggs.fredbloggs.fred@gmail.com > wrote: > >> On Sunday, February 16, 2014 9:35:42 PM UTC-5, John Larkin wrote: >> >>> >>> Nothing that you'd ever be involved with. >>> >> >> ...and you either from what I can see... > > I use sampling scopes all the time, and have a modest collection of antiques and > manuals. I have an actual Lumatron box > > https://dl.dropboxusercontent.com/u/53724080/Sampling/Lumatron.jpg > > (single point-contact diode sampler, avalanche transistor, hideous but > historical) > > and a Tek type N plugin (also single diode, but pretty) and a 1S1 and a 1S2 > > https://dl.dropboxusercontent.com/u/53724080/Sampling/1S2-547.jpg > > The 1S2 was a wonderful gadget; I learned a lot about picosecond electronics > from that plugin. > > A bunch of HP stuff, too > > https://dl.dropboxusercontent.com/u/53724080/Sampling/HP/1430guts.jpg > > https://dl.dropboxusercontent.com/u/53724080/Sampling/HP/HP185/188head3.jpg > > https://dl.dropboxusercontent.com/u/53724080/Sampling/HP/HP185/training2.jpg > > > On my bench I have a Tek 11802 with a bunch of sampling heads. The SD24 head > does 20 GHz sampling/TDR and, as Phil says, it makes a good picosecond pulse > generator, too. I recently got a 50 GHz head (SD32?) from ebay. Seems fine. > > I built a 2-diode sampler, similar to my sketch, and it worked fine. Tr was > about 70 ps, which is around 5 GHz. > > https://dl.dropboxusercontent.com/u/53724080/Sampling/Sampler1.JPG > > A 20 GHz sampler is feasible with surface-mount parts on FR4... I've seen it > done. It would be great fun to go into the sampler business, but it's hard to > compete with all the gear on ebay. > > Anybody who wants to get good at picosecond electronics can do it for $1000 and > some time. > > This was probably the world's first sampling oscilloscope: > > https://dl.dropboxusercontent.com/u/53724080/Sampling/ondo.gif > > > Here's Mark Kahr's superb paper on the history of sampling scopes: > > https://dl.dropboxusercontent.com/u/53724080/Sampling/MarkMTT.pdf
Interesting paper, thanks. It's roughly in IEEE format--Was it ever published in a journal, or was he just using the LaTeX style? Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
On Mon, 17 Feb 2014 10:49:14 -0800, Joerg <invalid@invalid.invalid> wrote:

>John Larkin wrote: >> On Mon, 17 Feb 2014 08:05:06 -0800, Joerg <invalid@invalid.invalid> wrote: >> >>> Phil Hobbs wrote: > >[...] > >>> >>>> In a fast sampler running inside a sampling loop (where the previous >>>> sample is added to the bias voltages), the diode conditions are kept >>>> pretty nearly constant from sample to sample. >>>> >>> Or one just lives with the delta-V without a loop. Could be averaged >>> locally or digitally later since AD converters have become so cheap. >> >> One problem with the single-diode sampler is that variations in the sampling >> pulse height rip right through with the signal, which can get noisy. And it >> tends to be nonlinear. And there is a huge kickout of the sampling pulse to the >> input connector. >> > >Yes, those are generally the downsides. But the sampling pulse can be >kept fairly constant and calculated out. The kickout is huge because the >sampling pulse roars right through the diode capacitance. But in >equivalent time samplers that is mostly just an EMC concern. Provided >that the load does not go berserk upon being hit with the kickout. > > >> The old Lumatron boxes, and the Tektronix N, were single-diode samplers with >> avalanche transistor strobe generators. They were fairly nasty. LeCroy did a >> single-diode sampler a while back but discontinued it. LeCroy did a lot of weird >> things, like their untriggered DDS timebase for telecom eye diagrams. >> > >Luckily my stuff doesn't need to be high precision but it has to be >cheap and small. For samplers it's truly slim pickens. Just now looking >for something useful for the post-sampler. All the commercial S&H chips >I saw so far are slow as snails, even the pricey ones. > > >> It would be interesting to see what could be done with MiniCircuits parts >> (baluns and mixers) to make an open-loop bridge sampler. >> > >I mostly came away disappointed when canvassing the commercial >offerings. That was the same during my masters project in the mid-80's. >I was designing a CCD camera that would beat the daylights out of the >camera the CCD manufacturer was offering (and it did). After a >frustrating search I just wound my own bridge sampler transformers. >Which I should have done in the first place, would have saved me time. >For the diodes I used matched quads, I believe from HP. Four individual >diodes per pouch.
Skyworks has a medium-barrier bridge quad, the SMS3930-021, and a high-barrier version, SMS3931. Both 0.3 to 0.5 pF per diode. Might work as an open-loop bridge sampler. -- John Larkin Highland Technology Inc www.highlandtechnology.com jlarkin at highlandtechnology dot com Precision electronic instrumentation
On Mon, 17 Feb 2014 13:59:32 -0500, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

>On 02/17/2014 01:22 PM, John Larkin wrote: >> On Mon, 17 Feb 2014 08:34:44 -0800 (PST), bloggs.fredbloggs.fred@gmail.com >> wrote: >> >>> On Sunday, February 16, 2014 9:35:42 PM UTC-5, John Larkin wrote: >>> >>>> >>>> Nothing that you'd ever be involved with. >>>> >>> >>> ...and you either from what I can see... >> >> I use sampling scopes all the time, and have a modest collection of antiques and >> manuals. I have an actual Lumatron box >> >> https://dl.dropboxusercontent.com/u/53724080/Sampling/Lumatron.jpg >> >> (single point-contact diode sampler, avalanche transistor, hideous but >> historical) >> >> and a Tek type N plugin (also single diode, but pretty) and a 1S1 and a 1S2 >> >> https://dl.dropboxusercontent.com/u/53724080/Sampling/1S2-547.jpg >> >> The 1S2 was a wonderful gadget; I learned a lot about picosecond electronics >> from that plugin. >> >> A bunch of HP stuff, too >> >> https://dl.dropboxusercontent.com/u/53724080/Sampling/HP/1430guts.jpg >> >> https://dl.dropboxusercontent.com/u/53724080/Sampling/HP/HP185/188head3.jpg >> >> https://dl.dropboxusercontent.com/u/53724080/Sampling/HP/HP185/training2.jpg >> >> >> On my bench I have a Tek 11802 with a bunch of sampling heads. The SD24 head >> does 20 GHz sampling/TDR and, as Phil says, it makes a good picosecond pulse >> generator, too. I recently got a 50 GHz head (SD32?) from ebay. Seems fine. >> >> I built a 2-diode sampler, similar to my sketch, and it worked fine. Tr was >> about 70 ps, which is around 5 GHz. >> >> https://dl.dropboxusercontent.com/u/53724080/Sampling/Sampler1.JPG >> >> A 20 GHz sampler is feasible with surface-mount parts on FR4... I've seen it >> done. It would be great fun to go into the sampler business, but it's hard to >> compete with all the gear on ebay. >> >> Anybody who wants to get good at picosecond electronics can do it for $1000 and >> some time. >> >> This was probably the world's first sampling oscilloscope: >> >> https://dl.dropboxusercontent.com/u/53724080/Sampling/ondo.gif >> >> >> Here's Mark Kahr's superb paper on the history of sampling scopes: >> >> https://dl.dropboxusercontent.com/u/53724080/Sampling/MarkMTT.pdf > >Interesting paper, thanks. It's roughly in IEEE format--Was it ever >published in a journal, or was he just using the LaTeX style? > >Cheers > >Phil Hobbs
I think he did it for an MTTS conference and it was published in the proceedings. You should meet Mark some day. He teaches EE at Pitt, and his wife is the dean of the CS department. He has an 11801, a gift from a friend. -- John Larkin Highland Technology Inc www.highlandtechnology.com jlarkin at highlandtechnology dot com Precision electronic instrumentation
On 02/17/2014 12:39 PM, Joerg wrote:
> Phil Hobbs wrote: >> On 02/17/2014 11:25 AM, Phil Hobbs wrote: >>> On 02/17/2014 11:05 AM, Joerg wrote: >>>> Phil Hobbs wrote: >>>>> On 02/17/2014 10:20 AM, Joerg wrote: >>>>>> Phil Hobbs wrote: >>>>>>> On 02/17/2014 01:55 AM, miso wrote: >>>>>>>> Phil Hobbs wrote: >>>>>>>> >>>>>>>>> On 2/15/2014 8:01 PM, Joerg wrote: >>>>>>>>>> John Larkin wrote: >>>>>>>>>>> On Sat, 15 Feb 2014 13:54:05 -0800, Joerg >>>>>>>>>>> <invalid@invalid.invalid> >>>>>>>>>>> wrote: >>>>>>>>>>> >>>>>>>>>>>> Folks, >>>>>>>>>>>> >>>>>>>>>>>> After John Larkin sent me SMS7621 Schottky RF diodes I played >>>>>>>>>>>> around >>>>>>>>>>>> with them in the lab. While they perform nicely at room >>>>>>>>>>>> temperature >>>>>>>>>>>> everything goes to pots at a slightly elevated temperature. At >>>>>>>>>>>> ballpark >>>>>>>>>>>> 50C the leakage current is already painfully high for use as a >>>>>>>>>>>> sampler >>>>>>>>>>>> diode. >>>>>>>>>>>> >>>>>>>>>>>> Question: Is there a fast RF diode or sampling diode in a >>>>>>>>>>>> reasonable >>>>>>>>>>>> cost range (low single-digit Dollars) with more barrier >>>>>>>>>>>> height? In >>>>>>>>>>>> samplers one doesn't care so much about Vf but unfortunately >>>>>>>>>>>> many >>>>>>>>>>>> such >>>>>>>>>>>> diodes are marketed as RF detectors where Vf does matter. I >>>>>>>>>>>> guess >>>>>>>>>>>> that's why they leak so much when the temps go up. >>>>>>>>>>> >>>>>>>>>>> The classic 2-diode feedback sampler doesn't much care about >>>>>>>>>>> leakage. >>>>>>>>>>> Within about a microsecond after the sampling pulse, the delta-v >>>>>>>>>>> glitch >>>>>>>>>>> is amplified in a charge amp, usually back up to 100% sampling >>>>>>>>>>> efficiency, and dumped into a slow s/h. That voltage, with the >>>>>>>>>>> diode >>>>>>>>>>> back-biases, is applied back onto the diodes for the next shot. >>>>>>>>>>> That >>>>>>>>>>> makes the overall sampling process ultra-linear and allows it to >>>>>>>>>>> work >>>>>>>>>>> down to very low trigger rates. >>>>>>>>>>> >>>>>>>>>> >>>>>>>>>> If push comes to shove I'll have to use a 2-diode sampler and >>>>>>>>>> figure >>>>>>>>>> out >>>>>>>>>> a balun for driving. I could also use a follow-up S&H on my >>>>>>>>>> single-diode >>>>>>>>>> sampler but it's a real estate and cost issue. Essentially my >>>>>>>>>> range >>>>>>>>>> gate >>>>>>>>>> will be around 300psec and the sampling cannot happen more than >>>>>>>>>> about >>>>>>>>>> once per usec. So around 3000:1. Delta-V is a few tens to hundreds >>>>>>>>>> of uV. >>>>>>>>>> >>>>>>>>>> >>>>>>>>>>> There are higher-barrier diodes in the Skyworks family. They >>>>>>>>>>> have a >>>>>>>>>>> great sample kit. >>>>>>>>>>> >>>>>>>>>> >>>>>>>>>> I think I'll have to talk to the guys there. Things like Schottky >>>>>>>>>> barrier height aren't in the datasheets but their FAEs should >>>>>>>>>> know. >>>>>>>>>> >>>>>>>>>> >>>>>>>>>>> What sort of risetime did you get? Are you doing an open-loop >>>>>>>>>>> sampler? >>>>>>>>>>> What the sampling pulse generator like? >>>>>>>>>>> >>>>>>>>>> >>>>>>>>>> I haven't tried the diodes in a sampler yet because the sampler >>>>>>>>>> only >>>>>>>>>> exists on paper so far. >>>>>>>>>> >>>>>>>>>> The sim shows about 120psec rise time. Open loop sampler. >>>>>>>>>> Essentially an >>>>>>>>>> RF BJT driven hard by a fast logic chip and then the collector >>>>>>>>>> signal is >>>>>>>>>> capacitively coupled into the sampler (fractions of a pF). >>>>>>>>>> >>>>>>>>>> One of the challenges with a post-S&H is that there seem to be >>>>>>>>>> no ICs >>>>>>>>>> for that. The SMP04 still needs 4usec, too long, and it costs an >>>>>>>>>> arm >>>>>>>>>> and >>>>>>>>>> a leg. It's a pity they don't sell the fast samplers from ADCs >>>>>>>>>> sans the >>>>>>>>>> ADC. "I'd like to have burger and fries but without the burger". >>>>>>>>>> >>>>>>>>> There are lots of nice analogue muxes with very low charge >>>>>>>>> injection, >>>>>>>>> and the dual-gate MOSFET trick works even better. >>>>>>>>> >>>>>>>>> Cheers >>>>>>>>> >>>>>>>>> Phil Hobbs >>>>>>>>> >>>>>>>> >>>>>>>> I'm all ears for the dual gate MOSFET trick. Are they using one of >>>>>>>> the >>>>>>>> gates >>>>>>>> to shield the charge injection of the other gate? >>>>>>>> >>>>>>>> Back in the day, much effort went into complimentary switches in >>>>>>>> charge >>>>>>>> transfer circuits. All sorts of games regarding clock rates (not too >>>>>>>> fast, >>>>>>>> not to slow), etc. >>>>>>>> >>>>>>>> >>>>>>> It's pretty simple--you connect the input to the source and G2, >>>>>>> storage >>>>>>> cap to D, sample pulse to G1. (It works best with parts that bias >>>>>>> properly with V_G2S = 0.) Works great with 3N201s. ;) >>>>>>> >>>>>>> You usually wouldn't do it that way because of the resistance >>>>>>> nonlinearity, but you don't care much about that in a sampler, >>>>>>> assuming >>>>>>> the sample gate is several time constants wide. >>>>>>> >>>>>> >>>>>> Unfortunately it hardly ever is. Mostly the time constant is 10x to >>>>>> 100x >>>>>> the sample gate width. So one either has to live just with the small >>>>>> delta-V as a sampled signal for processing or let this run over a >>>>>> hundred cycles or so and accumulate in a post sampler. Or use the >>>>>> feedback trick from the old HP days that John described to make the >>>>>> sampler cap ratchet up. >>>>>> >>>>>> The ratcheting should be easier with the dual-gate circuit though >>>>>> because there won't be any leakage. Might not even need the >>>>>> feedback or >>>>>> post-sampling at all. But capacitances are probably quite high, or at >>>>>> least higher than 0.25pF. >>>>>> >>>>>> Possibly the nonlinearityy could be servoed out by running a 2nd >>>>>> dual-gate in a slower fashion. That's how I sometimes did controllable >>>>>> delay circuits. It would be nice to have a dual-dual-gate FET for that >>>>>> but having them close to each other with some serious copper on the >>>>>> inner layers could work. >>>>>> >>>>> >>>>> Right, it's the second sampler I'm talking about. Normally getting a >>>>> 100-ns TC and a 1-us window isn't so hard. >>>>> >>>> >>>> Looks like I may have to live with single-diode and then there's only a >>>> few tens of nsec available. It might have to be discrete again since >>>> even what they call fast (ADG820 and similar) isn't that great in speed. >>>> At least in the reasonable price range. >>>> >>>> >>>>> In a fast sampler running inside a sampling loop (where the previous >>>>> sample is added to the bias voltages), the diode conditions are kept >>>>> pretty nearly constant from sample to sample. >>>>> >>>> >>>> Or one just lives with the delta-V without a loop. Could be averaged >>>> locally or digitally later since AD converters have become so cheap. >>>> >>> >>> Assuming that one has a sufficiently accurate model of diode behaviour >>> under dynamic conditions. Feedback is nice that way. >>> >>> A BF908 has a reverse transfer capacitance (G1-D) of typically 0.03 pF. >>> The G2-D capacitance would be more of a worry, but one more cascode >>> stage should fix that. >> >> Or run G2 from the buffer output, which would also fix it. Have to try >> that. >> > > True, but if you need a buffer or a charge amp with a beefy output you > might as well use a Schottky diode there, too :-) >
It only has to drive a 3-pF capacitance, and it's 3 pF to the input and not to the sample pulse. Plus when it's off, it's really off. But that part will run out of steam above a few hundred megahertz. My original suggestion was a cascode made out of two pHEMTs, e.g. SKY65050s. It would have much the same virtues at much lower capacitance, and has gain up to 12 GHz or so. The Skyworks ones are much nicer than the Avago ones for this sort of job, because their output conductance is a lot lower, i.e. they have a really high Early voltage. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
On Monday, February 17, 2014 1:22:26 PM UTC-5, John Larkin wrote:
> On Mon, 17 Feb 2014 08:34:44 -0800 (PST), bloggs.fredbloggs.fred@gmail.com > > wrote: > > > > >On Sunday, February 16, 2014 9:35:42 PM UTC-5, John Larkin wrote: > > > > > >> > > >> Nothing that you'd ever be involved with. > > >> > > > > > >...and you either from what I can see... > > > > I use sampling scopes all the time, and have a modest collection of antiques and > > manuals. I have an actual Lumatron box > > > > https://dl.dropboxusercontent.com/u/53724080/Sampling/Lumatron.jpg > > > > (single point-contact diode sampler, avalanche transistor, hideous but > > historical) > > > > and a Tek type N plugin (also single diode, but pretty) and a 1S1 and a 1S2 > > > > https://dl.dropboxusercontent.com/u/53724080/Sampling/1S2-547.jpg > > > > The 1S2 was a wonderful gadget; I learned a lot about picosecond electronics > > from that plugin. > > > > A bunch of HP stuff, too > > > > https://dl.dropboxusercontent.com/u/53724080/Sampling/HP/1430guts.jpg > > > > https://dl.dropboxusercontent.com/u/53724080/Sampling/HP/HP185/188head3.jpg > > > > https://dl.dropboxusercontent.com/u/53724080/Sampling/HP/HP185/training2.jpg > > > > > > On my bench I have a Tek 11802 with a bunch of sampling heads. The SD24 head > > does 20 GHz sampling/TDR and, as Phil says, it makes a good picosecond pulse > > generator, too. I recently got a 50 GHz head (SD32?) from ebay. Seems fine. > > > > I built a 2-diode sampler, similar to my sketch, and it worked fine. Tr was > > about 70 ps, which is around 5 GHz. > > > > https://dl.dropboxusercontent.com/u/53724080/Sampling/Sampler1.JPG > > > > A 20 GHz sampler is feasible with surface-mount parts on FR4... I've seen it > > done. It would be great fun to go into the sampler business, but it's hard to > > compete with all the gear on ebay. > > > > Anybody who wants to get good at picosecond electronics can do it for $1000 and > > some time. > > > > This was probably the world's first sampling oscilloscope: > > > > https://dl.dropboxusercontent.com/u/53724080/Sampling/ondo.gif > > > > > > Here's Mark Kahr's superb paper on the history of sampling scopes: > > > > https://dl.dropboxusercontent.com/u/53724080/Sampling/MarkMTT.pdf > > > > > > > > > > -- > > > > John Larkin Highland Technology Inc > > www.highlandtechnology.com jlarkin at highlandtechnology dot com > > > > Precision electronic instrumentation
That's like saying I'm "involved with" computer technology because I use a PC...
John Larkin wrote:
> On Mon, 17 Feb 2014 10:49:14 -0800, Joerg <invalid@invalid.invalid> wrote: > >> John Larkin wrote:
[...]
>> >>> It would be interesting to see what could be done with MiniCircuits parts >>> (baluns and mixers) to make an open-loop bridge sampler. >>> >> I mostly came away disappointed when canvassing the commercial >> offerings. That was the same during my masters project in the mid-80's. >> I was designing a CCD camera that would beat the daylights out of the >> camera the CCD manufacturer was offering (and it did). After a >> frustrating search I just wound my own bridge sampler transformers. >> Which I should have done in the first place, would have saved me time. >> For the diodes I used matched quads, I believe from HP. Four individual >> diodes per pouch. > > Skyworks has a medium-barrier bridge quad, the SMS3930-021, and a high-barrier > version, SMS3931. Both 0.3 to 0.5 pF per diode. Might work as an open-loop > bridge sampler. >
That may be better than the SMS7621 which also comes in a bridge quad. I'll call Skyworks this afternoon to see whether their higher barrier types are better in leakage current at higher temps. The board has to perform at least to 60C and I want to keep it simple, ideally without much fancy post-sampling and loop stuff. -- Regards, Joerg http://www.analogconsultants.com/
Phil Hobbs wrote:
> On 02/17/2014 12:39 PM, Joerg wrote: >> Phil Hobbs wrote: >>> On 02/17/2014 11:25 AM, Phil Hobbs wrote: >>>> On 02/17/2014 11:05 AM, Joerg wrote: >>>>> Phil Hobbs wrote: >>>>>> On 02/17/2014 10:20 AM, Joerg wrote: >>>>>>> Phil Hobbs wrote: >>>>>>>> On 02/17/2014 01:55 AM, miso wrote: >>>>>>>>> Phil Hobbs wrote: >>>>>>>>> >>>>>>>>>> On 2/15/2014 8:01 PM, Joerg wrote: >>>>>>>>>>> John Larkin wrote: >>>>>>>>>>>> On Sat, 15 Feb 2014 13:54:05 -0800, Joerg >>>>>>>>>>>> <invalid@invalid.invalid> >>>>>>>>>>>> wrote: >>>>>>>>>>>> >>>>>>>>>>>>> Folks, >>>>>>>>>>>>> >>>>>>>>>>>>> After John Larkin sent me SMS7621 Schottky RF diodes I played >>>>>>>>>>>>> around >>>>>>>>>>>>> with them in the lab. While they perform nicely at room >>>>>>>>>>>>> temperature >>>>>>>>>>>>> everything goes to pots at a slightly elevated temperature. At >>>>>>>>>>>>> ballpark >>>>>>>>>>>>> 50C the leakage current is already painfully high for use as a >>>>>>>>>>>>> sampler >>>>>>>>>>>>> diode. >>>>>>>>>>>>> >>>>>>>>>>>>> Question: Is there a fast RF diode or sampling diode in a >>>>>>>>>>>>> reasonable >>>>>>>>>>>>> cost range (low single-digit Dollars) with more barrier >>>>>>>>>>>>> height? In >>>>>>>>>>>>> samplers one doesn't care so much about Vf but unfortunately >>>>>>>>>>>>> many >>>>>>>>>>>>> such >>>>>>>>>>>>> diodes are marketed as RF detectors where Vf does matter. I >>>>>>>>>>>>> guess >>>>>>>>>>>>> that's why they leak so much when the temps go up. >>>>>>>>>>>> >>>>>>>>>>>> The classic 2-diode feedback sampler doesn't much care about >>>>>>>>>>>> leakage. >>>>>>>>>>>> Within about a microsecond after the sampling pulse, the >>>>>>>>>>>> delta-v >>>>>>>>>>>> glitch >>>>>>>>>>>> is amplified in a charge amp, usually back up to 100% sampling >>>>>>>>>>>> efficiency, and dumped into a slow s/h. That voltage, with the >>>>>>>>>>>> diode >>>>>>>>>>>> back-biases, is applied back onto the diodes for the next shot. >>>>>>>>>>>> That >>>>>>>>>>>> makes the overall sampling process ultra-linear and allows >>>>>>>>>>>> it to >>>>>>>>>>>> work >>>>>>>>>>>> down to very low trigger rates. >>>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> If push comes to shove I'll have to use a 2-diode sampler and >>>>>>>>>>> figure >>>>>>>>>>> out >>>>>>>>>>> a balun for driving. I could also use a follow-up S&H on my >>>>>>>>>>> single-diode >>>>>>>>>>> sampler but it's a real estate and cost issue. Essentially my >>>>>>>>>>> range >>>>>>>>>>> gate >>>>>>>>>>> will be around 300psec and the sampling cannot happen more than >>>>>>>>>>> about >>>>>>>>>>> once per usec. So around 3000:1. Delta-V is a few tens to >>>>>>>>>>> hundreds >>>>>>>>>>> of uV. >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>>> There are higher-barrier diodes in the Skyworks family. They >>>>>>>>>>>> have a >>>>>>>>>>>> great sample kit. >>>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> I think I'll have to talk to the guys there. Things like >>>>>>>>>>> Schottky >>>>>>>>>>> barrier height aren't in the datasheets but their FAEs should >>>>>>>>>>> know. >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>>> What sort of risetime did you get? Are you doing an open-loop >>>>>>>>>>>> sampler? >>>>>>>>>>>> What the sampling pulse generator like? >>>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> I haven't tried the diodes in a sampler yet because the sampler >>>>>>>>>>> only >>>>>>>>>>> exists on paper so far. >>>>>>>>>>> >>>>>>>>>>> The sim shows about 120psec rise time. Open loop sampler. >>>>>>>>>>> Essentially an >>>>>>>>>>> RF BJT driven hard by a fast logic chip and then the collector >>>>>>>>>>> signal is >>>>>>>>>>> capacitively coupled into the sampler (fractions of a pF). >>>>>>>>>>> >>>>>>>>>>> One of the challenges with a post-S&H is that there seem to be >>>>>>>>>>> no ICs >>>>>>>>>>> for that. The SMP04 still needs 4usec, too long, and it costs an >>>>>>>>>>> arm >>>>>>>>>>> and >>>>>>>>>>> a leg. It's a pity they don't sell the fast samplers from ADCs >>>>>>>>>>> sans the >>>>>>>>>>> ADC. "I'd like to have burger and fries but without the burger". >>>>>>>>>>> >>>>>>>>>> There are lots of nice analogue muxes with very low charge >>>>>>>>>> injection, >>>>>>>>>> and the dual-gate MOSFET trick works even better. >>>>>>>>>> >>>>>>>>>> Cheers >>>>>>>>>> >>>>>>>>>> Phil Hobbs >>>>>>>>>> >>>>>>>>> >>>>>>>>> I'm all ears for the dual gate MOSFET trick. Are they using one of >>>>>>>>> the >>>>>>>>> gates >>>>>>>>> to shield the charge injection of the other gate? >>>>>>>>> >>>>>>>>> Back in the day, much effort went into complimentary switches in >>>>>>>>> charge >>>>>>>>> transfer circuits. All sorts of games regarding clock rates >>>>>>>>> (not too >>>>>>>>> fast, >>>>>>>>> not to slow), etc. >>>>>>>>> >>>>>>>>> >>>>>>>> It's pretty simple--you connect the input to the source and G2, >>>>>>>> storage >>>>>>>> cap to D, sample pulse to G1. (It works best with parts that bias >>>>>>>> properly with V_G2S = 0.) Works great with 3N201s. ;) >>>>>>>> >>>>>>>> You usually wouldn't do it that way because of the resistance >>>>>>>> nonlinearity, but you don't care much about that in a sampler, >>>>>>>> assuming >>>>>>>> the sample gate is several time constants wide. >>>>>>>> >>>>>>> >>>>>>> Unfortunately it hardly ever is. Mostly the time constant is 10x to >>>>>>> 100x >>>>>>> the sample gate width. So one either has to live just with the small >>>>>>> delta-V as a sampled signal for processing or let this run over a >>>>>>> hundred cycles or so and accumulate in a post sampler. Or use the >>>>>>> feedback trick from the old HP days that John described to make the >>>>>>> sampler cap ratchet up. >>>>>>> >>>>>>> The ratcheting should be easier with the dual-gate circuit though >>>>>>> because there won't be any leakage. Might not even need the >>>>>>> feedback or >>>>>>> post-sampling at all. But capacitances are probably quite high, >>>>>>> or at >>>>>>> least higher than 0.25pF. >>>>>>> >>>>>>> Possibly the nonlinearityy could be servoed out by running a 2nd >>>>>>> dual-gate in a slower fashion. That's how I sometimes did >>>>>>> controllable >>>>>>> delay circuits. It would be nice to have a dual-dual-gate FET for >>>>>>> that >>>>>>> but having them close to each other with some serious copper on the >>>>>>> inner layers could work. >>>>>>> >>>>>> >>>>>> Right, it's the second sampler I'm talking about. Normally getting a >>>>>> 100-ns TC and a 1-us window isn't so hard. >>>>>> >>>>> >>>>> Looks like I may have to live with single-diode and then there's >>>>> only a >>>>> few tens of nsec available. It might have to be discrete again since >>>>> even what they call fast (ADG820 and similar) isn't that great in >>>>> speed. >>>>> At least in the reasonable price range. >>>>> >>>>> >>>>>> In a fast sampler running inside a sampling loop (where the previous >>>>>> sample is added to the bias voltages), the diode conditions are kept >>>>>> pretty nearly constant from sample to sample. >>>>>> >>>>> >>>>> Or one just lives with the delta-V without a loop. Could be averaged >>>>> locally or digitally later since AD converters have become so cheap. >>>>> >>>> >>>> Assuming that one has a sufficiently accurate model of diode behaviour >>>> under dynamic conditions. Feedback is nice that way. >>>> >>>> A BF908 has a reverse transfer capacitance (G1-D) of typically 0.03 pF. >>>> The G2-D capacitance would be more of a worry, but one more cascode >>>> stage should fix that. >>> >>> Or run G2 from the buffer output, which would also fix it. Have to try >>> that. >>> >> >> True, but if you need a buffer or a charge amp with a beefy output you >> might as well use a Schottky diode there, too :-) >> > > It only has to drive a 3-pF capacitance, and it's 3 pF to the input and > not to the sample pulse. Plus when it's off, it's really off. But that > part will run out of steam above a few hundred megahertz. > > My original suggestion was a cascode made out of two pHEMTs, e.g. > SKY65050s. It would have much the same virtues at much lower > capacitance, and has gain up to 12 GHz or so. The Skyworks ones are > much nicer than the Avago ones for this sort of job, because their > output conductance is a lot lower, i.e. they have a really high Early > voltage. >
They are nice but most likely they'd have to be adjusted in production for the upper gate bias (where the to-be-sampled signals would go in). The datasheet mentions about 30% total tolrance in the pinchoff. I guess I could automate that. The other thing is whether it'll be stable with one riding on the drain of the other. With GHz devices that can be like trying to balance a garden hose on the tip of a finger. Wish they had a SPICE model for the SKY65050, there's only ADS and S-Parameters on their site. http://www.skyworksinc.com/Product.aspx?ProductID=348 -- Regards, Joerg http://www.analogconsultants.com/
On 2/16/2014 9:39 PM, John Larkin wrote:
> On Mon, 17 Feb 2014 02:50:42 +0100, Gerhard Hoffmann <dk4xp@arcor.de> wrote: > >> Am 17.02.2014 00:45, schrieb Joerg: >>> John Larkin wrote: >>>> I've done a little playing with software deconvolution for TDR. The idea is to >>>> build a fast but ugly sampling or TDR system and clean up its step response with >>>> a software FIR filter, the problem being how to train the filter. >>>> >>>> https://dl.dropboxusercontent.com/u/53724080/Sampling/TDR_Decon_demo.jpg >>>> >>>> I have an algorithm that seems to work well, although I've never really >>>> understood it. It just seemed like a reasonable thing to do. >> >> HP had the deconvolution software in their 54750A scopes & >> associated TDRs. There is a PSPL app note that compares >> industry standard samplers that talks about that. >> >> regards, Gerhard >> > > I think that one used the Bracewell Transform, sort of an FFT that is its own > inverse. > > The classic deconvolution was an FFT-divide-reverseFFT, since convolution in the > time domain is multiplication in the frequency domain. But the FFT thing has > noise and divide-by-zero problems. Imagine a graphic equalized where you have to > crank some of the gains to infinity, to make zero bins come out flat. > >
Bracewell's version, which he called the Fast Hartley Transform, is equivalent to the FFT for real-valued signals and saves half the arithmetic. That mattered more in 1983 than it does now, but it's still nice, once you get to the last couple of turns of the optimization crank. Bracewell was one of my favourite professors--extremely lucid and down-to-earth, plus he cared a good deal more for his students than for his ego. His class was one of the two most useful ones I took in grad school. (The other one was a class in asymptotic methods, taught by an EE-turned-matho, Stefanos Venakides. He didn't get tenure and had to move to Brown, where I think he did fine.) But you have to admit that Agoston's version is pure art. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
On Mon, 17 Feb 2014 18:22:27 -0500, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

>On 2/16/2014 9:39 PM, John Larkin wrote: >> On Mon, 17 Feb 2014 02:50:42 +0100, Gerhard Hoffmann <dk4xp@arcor.de> wrote: >> >>> Am 17.02.2014 00:45, schrieb Joerg: >>>> John Larkin wrote: >>>>> I've done a little playing with software deconvolution for TDR. The idea is to >>>>> build a fast but ugly sampling or TDR system and clean up its step response with >>>>> a software FIR filter, the problem being how to train the filter. >>>>> >>>>> https://dl.dropboxusercontent.com/u/53724080/Sampling/TDR_Decon_demo.jpg >>>>> >>>>> I have an algorithm that seems to work well, although I've never really >>>>> understood it. It just seemed like a reasonable thing to do. >>> >>> HP had the deconvolution software in their 54750A scopes & >>> associated TDRs. There is a PSPL app note that compares >>> industry standard samplers that talks about that. >>> >>> regards, Gerhard >>> >> >> I think that one used the Bracewell Transform, sort of an FFT that is its own >> inverse. >> >> The classic deconvolution was an FFT-divide-reverseFFT, since convolution in the >> time domain is multiplication in the frequency domain. But the FFT thing has >> noise and divide-by-zero problems. Imagine a graphic equalized where you have to >> crank some of the gains to infinity, to make zero bins come out flat. >> >> > >Bracewell's version, which he called the Fast Hartley Transform, is >equivalent to the FFT for real-valued signals and saves half the >arithmetic. That mattered more in 1983 than it does now, but it's still >nice, once you get to the last couple of turns of the optimization crank. > >Bracewell was one of my favourite professors--extremely lucid and >down-to-earth, plus he cared a good deal more for his students than for >his ego. His class was one of the two most useful ones I took in grad >school. (The other one was a class in asymptotic methods, taught by an >EE-turned-matho, Stefanos Venakides. He didn't get tenure and had to >move to Brown, where I think he did fine.) > >But you have to admit that Agoston's version is pure art. > >Cheers > >Phil Hobbs
If you look at Agoston's Tek samplers, it's mind-boggling to think about how many hours they had to have spent to get it as clean as it is. There are strange wirebonds all over the place, notched microstrips, bizarre shapes, scary stuff. It must have been a nightmare in production and test, too. https://dl.dropboxusercontent.com/u/53724080/SD-24/SD-24.zip Agoston's original SD24-looking patent used a trench/slotline machined into a metal block as the sampling pulse shaper and transmission line. I wonder why they didn't do it that way. He emailed me a couple years ago about my deconvolution thing, and I sent him the demo and the source code. Don't know if he ever used it. -- John Larkin Highland Technology, Inc jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
Den tirsdag den 18. februar 2014 00.44.55 UTC+1 skrev John Larkin:
> On Mon, 17 Feb 2014 18:22:27 -0500, Phil Hobbs > > <pcdhSpamMeSenseless@electrooptical.net> wrote: > > > > >On 2/16/2014 9:39 PM, John Larkin wrote: > > >> On Mon, 17 Feb 2014 02:50:42 +0100, Gerhard Hoffmann <dk4xp@arcor.de> wrote: > > >> > > >>> Am 17.02.2014 00:45, schrieb Joerg: > > >>>> John Larkin wrote: > > >>>>> I've done a little playing with software deconvolution for TDR. The idea is to > > >>>>> build a fast but ugly sampling or TDR system and clean up its step response with > > >>>>> a software FIR filter, the problem being how to train the filter. > > >>>>> > > >>>>> https://dl.dropboxusercontent.com/u/53724080/Sampling/TDR_Decon_demo.jpg > > >>>>> > > >>>>> I have an algorithm that seems to work well, although I've never really > > >>>>> understood it. It just seemed like a reasonable thing to do. > > >>> > > >>> HP had the deconvolution software in their 54750A scopes & > > >>> associated TDRs. There is a PSPL app note that compares > > >>> industry standard samplers that talks about that. > > >>> > > >>> regards, Gerhard > > >>> > > >> > > >> I think that one used the Bracewell Transform, sort of an FFT that is its own > > >> inverse. > > >> > > >> The classic deconvolution was an FFT-divide-reverseFFT, since convolution in the > > >> time domain is multiplication in the frequency domain. But the FFT thing has > > >> noise and divide-by-zero problems. Imagine a graphic equalized where you have to > > >> crank some of the gains to infinity, to make zero bins come out flat. > > >> > > >> > > > > > >Bracewell's version, which he called the Fast Hartley Transform, is > > >equivalent to the FFT for real-valued signals and saves half the > > >arithmetic. That mattered more in 1983 than it does now, but it's still > > >nice, once you get to the last couple of turns of the optimization crank. > > > > > >Bracewell was one of my favourite professors--extremely lucid and > > >down-to-earth, plus he cared a good deal more for his students than for > > >his ego. His class was one of the two most useful ones I took in grad > > >school. (The other one was a class in asymptotic methods, taught by an > > >EE-turned-matho, Stefanos Venakides. He didn't get tenure and had to > > >move to Brown, where I think he did fine.) > > > > > >But you have to admit that Agoston's version is pure art. > > > > > >Cheers > > > > > >Phil Hobbs > > > > If you look at Agoston's Tek samplers, it's mind-boggling to think > > about how many hours they had to have spent to get it as clean as it > > is. There are strange wirebonds all over the place, notched > > microstrips, bizarre shapes, scary stuff. It must have been a > > nightmare in production and test, too. > > > > https://dl.dropboxusercontent.com/u/53724080/SD-24/SD-24.zip >
I visited Rohde & Schwarz in Munich many years ago and saw the place where they made stuff like that the guy there had a German saying something like: the higher the frequency the longer the face -Lasse