Post by thread:Very Fast RAM

I am looking for very fast ram/fifo buffer ICs for a possible digital oscilloscope project. Mainly, I am wondering what the fastest commonly available RAM or FIFO is, and how fast it is. By doing a search on the web, the fastest I could find was about 3.3 or 4 ns access time. This would probably work, but I am just wondering what exists. Does anyone know what DSOs do to capture data at >2GS/s ?? I thought possibly they might parallel several 5ns FSRAMs and let one recover while writing to the next, etc. There still is a limit to how much one could do this, thought, because the data has to be placed at the pins of the ram for a certain amt. of time. Thanks, Sean +--------------------------------+ | Sean Breheny | | Amateur Radio Callsign: KA3YXM | | Electrical Engineering Student | +--------------------------------+ http://www.people.cornell.edu/pages/shb7 spam_OUTshb7TakeThisOuTcornell.edu Phone(USA): (607) 253-0315

At 01:00 PM 12/12/97 +1000, you wrote: >As far as I know, these fast CRO's, (if you can call that anymore), >sample a repetitive wave form over and over and build up a picture of >it by shifting the trigger point slightly on each scan. So in actual fact they {Quote hidden}> are >not sampling the wave form at GHz rates, maybe down to KHz rates. > >I haven't kept up with the latest gear though, so I may be out of >date here. > > >Tony. > > >Just when I thought I knew it all, >I learned that I didn't. > I was just looking at a few oscilloscope data sheets and, you are right, many of these scopes sample at repetitive rates to achieve their peak performance. However, several can do flat-out 1GS/s rates, the HP Infinium, is one example. Sean +--------------------------------+ | Sean Breheny | | Amateur Radio Callsign: KA3YXM | | Electrical Engineering Student | +--------------------------------+ http://www.people.cornell.edu/pages/shb7 .....shb7KILLspam@spam@cornell.edu Phone(USA): (607) 253-0315

Sean, I've had a DSO project on the `back burner' for a long time. I recently had a requirement to analyze IR remote signals so I'm going to finish the project. Coincidentally, the recent issue of Circuit Cellar Ink (#89 Dec 97) has two logic analyzer projects. One is a ram-based 50Mhz design and the other is a FIFO-based 20Mhz design. Both connect to a bidirectional PC parallel port. My design will be PIC-based and connect to the serial port. - Tom At 07:47 PM 12/11/97 -0500, you wrote: {Quote hidden}>I am looking for very fast ram/fifo buffer ICs for a possible digital >oscilloscope project. Mainly, I am wondering what the fastest commonly >available RAM or FIFO is, and how fast it is. By doing a search on the web, >the fastest I could find was about 3.3 or 4 ns access time. This would >probably work, but I am just wondering what exists. Does anyone know what >DSOs do to capture data at >2GS/s ?? I thought possibly they might parallel >several 5ns FSRAMs and let one recover while writing to the next, etc. >There still is a limit to how much one could do this, thought, because the >data has to be placed at the pins of the ram for a certain amt. of time. > >Thanks, > >Sean

>I was just looking at a few oscilloscope data sheets and, you are right, >many of these scopes sample at repetitive rates to achieve their peak >performance. However, several can do flat-out 1GS/s rates, the HP Infinium, >is one example. Yep, some can, some can't. Another technique used in some of the older digital scopes was to clock the analog data into a fast ccd, and then clock it out to an adc at a much slower rate. Philips and Tek both used this method, but they also had the ability to custom manufacture the required devices. Tek also made some digitizers using a video-camera like device that sweeped an electron beam across a 512 * 512 pixel diode target. The charge on each 'pixel' of the target could then be read out after the trace and either displayed (analog) or digitized and stored. This topped out at something like the equivalent of 500 GSamples/second! As far as I know, all the newer scopes have gone to using fast flash ADCs. Seems to me the problem for the individual builder isn't finding fast ram (after all, you could interleave 16 sets of common 15 ns cache sram), but the front end and adc circuits. Even a 300-400 MHz analog bandwidth is going to require some pretty 'tweaky' amps and attenuators, and lots o' gain-bandwidth product in the active devices. Very fast flash converters aren't likely to be inexpensive, and their high (and variable) input capacitance isn't trivial to drive. Good luck, newell

At 09:16 PM 12/11/97 -0600, you wrote: {Quote hidden}>>I was just looking at a few oscilloscope data sheets and, you are right, >>many of these scopes sample at repetitive rates to achieve their peak >>performance. However, several can do flat-out 1GS/s rates, the HP Infinium, >>is one example. > >Yep, some can, some can't. Another technique used in some of the older >digital scopes was to clock the analog data into a fast ccd, and then clock >it out to an adc at a much slower rate. Philips and Tek both used this >method, but they also had the ability to custom manufacture the required >devices. > >Tek also made some digitizers using a video-camera like device that sweeped >an electron beam across a 512 * 512 pixel diode target. The charge on each >'pixel' of the target could then be read out after the trace and either >displayed (analog) or digitized and stored. This topped out at something >like the equivalent of 500 GSamples/second! > >As far as I know, all the newer scopes have gone to using fast flash ADCs. > >Seems to me the problem for the individual builder isn't finding fast ram >(after all, you could interleave 16 sets of common 15 ns cache sram), but >the front end and adc circuits. Even a 300-400 MHz analog bandwidth is >going to require some pretty 'tweaky' amps and attenuators, and lots o' >gain-bandwidth product in the active devices. Very fast flash converters >aren't likely to be inexpensive, and their high (and variable) input >capacitance isn't trivial to drive. > Well, my idea is somewhere in the neighborhood of 100MHz analog BW. I realize that I don't have that much design experience to get good results up to several 100s of MHz., but I don't see why 100MHz would be too hard. There are lots of video amps and 100MHz op amps out there which are not too expensive and (so it would seem to me) have BW around 100MHz. I have found 30MS/s ADCs for $10. I hope, and I don't see why it would be too hard, to buffer the input of this ADC with a 100MHz op amp, a (roughly) purely resistive attenuator (made from a rotary switch and several resistors), send the output of the ADC to some SRAM, clock the ADC with a PLL with programmable dividing radio (to get adjustable sample rate with very accurate time base), read the output either to a PC or to a microcontroller (PIC :-) ) at a slower rate to feed to an LCD. I could use the multiple sampling technique to digitize at more than 30MS/s as well as go flat out at 30MS/s for up to 15MHz non-periodic stuff. I realize that this would not be as good as a $1000 Fluke scopemeter or similar device, but for my budget, I think it would make an interesting project and be able to visualize waveforms for most hobby/student applications. Am I wrong/wasting my money? Thanks for the help, Sean +--------------------------------+ | Sean Breheny | | Amateur Radio Callsign: KA3YXM | | Electrical Engineering Student | +--------------------------------+ http://www.people.cornell.edu/pages/shb7 shb7KILLspamcornell.edu Phone(USA): (607) 253-0315

If you read 'Analogue design, art science and personalities' (I think this is the name) you'll see quite a bit of material on how TEK designed high-speed vertical amplifiers. {Quote hidden}> ---------- > From: Sean Breheny[SMTP:.....shb7KILLspam.....CORNELL.EDU] > Reply To: pic microcontroller discussion list > Sent: Friday, December 12, 1997 4:36 PM > To: EraseMEPICLISTspam_OUTTakeThisOuTMITVMA.MIT.EDU > Subject: Re: Very Fast RAM > > At 09:16 PM 12/11/97 -0600, you wrote: > >>I was just looking at a few oscilloscope data sheets and, you are > right, > >>many of these scopes sample at repetitive rates to achieve their > peak > >>performance. However, several can do flat-out 1GS/s rates, the HP > Infinium, > >>is one example. > > > >Yep, some can, some can't. Another technique used in some of the > older > >digital scopes was to clock the analog data into a fast ccd, and then > clock > >it out to an adc at a much slower rate. Philips and Tek both used > this > >method, but they also had the ability to custom manufacture the > required > >devices. > > > >Tek also made some digitizers using a video-camera like device that > sweeped > >an electron beam across a 512 * 512 pixel diode target. The charge > on each > >'pixel' of the target could then be read out after the trace and > either > >displayed (analog) or digitized and stored. This topped out at > something > >like the equivalent of 500 GSamples/second! > > > >As far as I know, all the newer scopes have gone to using fast flash > ADCs. > > > >Seems to me the problem for the individual builder isn't finding fast > ram > >(after all, you could interleave 16 sets of common 15 ns cache sram), > but > >the front end and adc circuits. Even a 300-400 MHz analog bandwidth > is > >going to require some pretty 'tweaky' amps and attenuators, and lots > o' > >gain-bandwidth product in the active devices. Very fast flash > converters > >aren't likely to be inexpensive, and their high (and variable) input > >capacitance isn't trivial to drive. > > > > Well, my idea is somewhere in the neighborhood of 100MHz analog BW. I > realize that I don't have that much design experience to get good > results > up to several 100s of MHz., but I don't see why 100MHz would be too > hard. > There are lots of video amps and 100MHz op amps out there which are > not too > expensive and (so it would seem to me) have BW around 100MHz. I have > found > 30MS/s ADCs for $10. I hope, and I don't see why it would be too hard, > to > buffer the input of this ADC with a 100MHz op amp, a (roughly) purely > resistive attenuator (made from a rotary switch and several > resistors), > send the output of the ADC to some SRAM, clock the ADC with a PLL with > programmable dividing radio (to get adjustable sample rate with very > accurate time base), read the output either to a PC or to a > microcontroller > (PIC :-) ) at a slower rate to feed to an LCD. I could use the > multiple > sampling technique to digitize at more than 30MS/s as well as go flat > out > at 30MS/s for up to 15MHz non-periodic stuff. I realize that this > would not > be as good as a $1000 Fluke scopemeter or similar device, but for my > budget, I think it would make an interesting project and be able to > visualize waveforms for most hobby/student applications. Am I > wrong/wasting > my money? > > > Thanks for the help, > > Sean > +--------------------------------+ > | Sean Breheny | > | Amateur Radio Callsign: KA3YXM | > | Electrical Engineering Student | > +--------------------------------+ > http://www.people.cornell.edu/pages/shb7 > shb7spam_OUTcornell.edu > Phone(USA): (607) 253-0315 >

yes, but couldn't you interleave A/Ds as well. i.e each A/D handles the nth sample?. In front of the A/Ds we have a bunch of high-speed sample and hold circuits, and we route the input signal to the appropriate A/D in a 'round-robin' fashion. Our friends at Los Alamos would have needed this sort of thing - as someone at TEK said 'our customers needed scopes which would capture events with a repetition rate of one every month, or so'. {Quote hidden}> ---------- > From: Scott Newell[SMTP:@spam@newellKILLspamCEI.NET] > Reply To: pic microcontroller discussion list > Sent: Friday, December 12, 1997 4:16 PM > To: KILLspamPICLISTKILLspamMITVMA.MIT.EDU > Subject: Re: Very Fast RAM > > >I was just looking at a few oscilloscope data sheets and, you are > right, > >many of these scopes sample at repetitive rates to achieve their peak > >performance. However, several can do flat-out 1GS/s rates, the HP > Infinium, > >is one example. > > Yep, some can, some can't. Another technique used in some of the > older > digital scopes was to clock the analog data into a fast ccd, and then > clock > it out to an adc at a much slower rate. Philips and Tek both used > this > method, but they also had the ability to custom manufacture the > required > devices. > > Tek also made some digitizers using a video-camera like device that > sweeped > an electron beam across a 512 * 512 pixel diode target. The charge on > each > 'pixel' of the target could then be read out after the trace and > either > displayed (analog) or digitized and stored. This topped out at > something > like the equivalent of 500 GSamples/second! > > As far as I know, all the newer scopes have gone to using fast flash > ADCs. > > Seems to me the problem for the individual builder isn't finding fast > ram > (after all, you could interleave 16 sets of common 15 ns cache sram), > but > the front end and adc circuits. Even a 300-400 MHz analog bandwidth > is > going to require some pretty 'tweaky' amps and attenuators, and lots > o' > gain-bandwidth product in the active devices. Very fast flash > converters > aren't likely to be inexpensive, and their high (and variable) input > capacitance isn't trivial to drive. > > > Good luck, > newell >

At 05:29 PM 12/12/97 +1300, you wrote: >If you read 'Analogue design, art science and personalities' (I think >this is the name) you'll see quite a bit of material on how TEK designed >high-speed vertical amplifiers. > > Thanks for the tip. I am still wondering, though, what in general do people think of my idea? What am I forgetting? Any suggestions? Is it worth it? Thanks, Sean {Quote hidden}>> ---------- >> From: Sean Breheny[SMTP:RemoveMEshb7TakeThisOuTCORNELL.EDU] >> Reply To: pic microcontroller discussion list >> Sent: Friday, December 12, 1997 4:36 PM >> To: spamBeGonePICLISTspamBeGoneMITVMA.MIT.EDU >> Subject: Re: Very Fast RAM >> >> At 09:16 PM 12/11/97 -0600, you wrote: >> >>I was just looking at a few oscilloscope data sheets and, you are >> right, >> >>many of these scopes sample at repetitive rates to achieve their >> peak >> >>performance. However, several can do flat-out 1GS/s rates, the HP >> Infinium, >> >>is one example. >> > >> >Yep, some can, some can't. Another technique used in some of the >> older >> >digital scopes was to clock the analog data into a fast ccd, and then >> clock >> >it out to an adc at a much slower rate. Philips and Tek both used >> this >> >method, but they also had the ability to custom manufacture the >> required >> >devices. >> > >> >Tek also made some digitizers using a video-camera like device that >> sweeped >> >an electron beam across a 512 * 512 pixel diode target. The charge >> on each >> >'pixel' of the target could then be read out after the trace and >> either >> >displayed (analog) or digitized and stored. This topped out at >> something >> >like the equivalent of 500 GSamples/second! >> > >> >As far as I know, all the newer scopes have gone to using fast flash >> ADCs. >> > >> >Seems to me the problem for the individual builder isn't finding fast >> ram >> >(after all, you could interleave 16 sets of common 15 ns cache sram), >> but >> >the front end and adc circuits. Even a 300-400 MHz analog bandwidth >> is >> >going to require some pretty 'tweaky' amps and attenuators, and lots >> o' >> >gain-bandwidth product in the active devices. Very fast flash >> converters >> >aren't likely to be inexpensive, and their high (and variable) input >> >capacitance isn't trivial to drive. >> > >> >> Well, my idea is somewhere in the neighborhood of 100MHz analog BW. I >> realize that I don't have that much design experience to get good >> results >> up to several 100s of MHz., but I don't see why 100MHz would be too >> hard. >> There are lots of video amps and 100MHz op amps out there which are >> not too >> expensive and (so it would seem to me) have BW around 100MHz. I have >> found >> 30MS/s ADCs for $10. I hope, and I don't see why it would be too hard, >> to >> buffer the input of this ADC with a 100MHz op amp, a (roughly) purely >> resistive attenuator (made from a rotary switch and several >> resistors), >> send the output of the ADC to some SRAM, clock the ADC with a PLL with >> programmable dividing radio (to get adjustable sample rate with very >> accurate time base), read the output either to a PC or to a >> microcontroller >> (PIC :-) ) at a slower rate to feed to an LCD. I could use the >> multiple >> sampling technique to digitize at more than 30MS/s as well as go flat >> out >> at 30MS/s for up to 15MHz non-periodic stuff. I realize that this >> would not >> be as good as a $1000 Fluke scopemeter or similar device, but for my >> budget, I think it would make an interesting project and be able to >> visualize waveforms for most hobby/student applications. Am I >> wrong/wasting >> my money? >> >> >> Thanks for the help, >> >> Sean >> +--------------------------------+ >> | Sean Breheny | >> | Amateur Radio Callsign: KA3YXM | >> | Electrical Engineering Student | >> +--------------------------------+ >> http://www.people.cornell.edu/pages/shb7 >> TakeThisOuTshb7EraseMEspam_OUTcornell.edu >> Phone(USA): (607) 253-0315 >> > +--------------------------------+ | Sean Breheny | | Amateur Radio Callsign: KA3YXM | | Electrical Engineering Student | +--------------------------------+ http://www.people.cornell.edu/pages/shb7 RemoveMEshb7TakeThisOuTcornell.edu Phone(USA): (607) 253-0315

Scott Newell wrote: {Quote hidden}> > >I was just looking at a few oscilloscope data sheets and, you are right, > >many of these scopes sample at repetitive rates to achieve their peak > >performance. However, several can do flat-out 1GS/s rates, the HP Infinium, > >is one example. > > Yep, some can, some can't. Another technique used in some of the older > digital scopes was to clock the analog data into a fast ccd, and then clock > it out to an adc at a much slower rate. Philips and Tek both used this > method, but they also had the ability to custom manufacture the required > devices. > > Tek also made some digitizers using a video-camera like device that sweeped > an electron beam across a 512 * 512 pixel diode target. The charge on each > 'pixel' of the target could then be read out after the trace and either > displayed (analog) or digitized and stored. This topped out at something > like the equivalent of 500 GSamples/second! > > As far as I know, all the newer scopes have gone to using fast flash ADCs. > > Seems to me the problem for the individual builder isn't finding fast ram > (after all, you could interleave 16 sets of common 15 ns cache sram), but > the front end and adc circuits. Even a 300-400 MHz analog bandwidth is > going to require some pretty 'tweaky' amps and attenuators, and lots o' > gain-bandwidth product in the active devices. Very fast flash converters > aren't likely to be inexpensive, and their high (and variable) input > capacitance isn't trivial to drive. A nice trick if you want to fill up memory fast, is to use a PR counter on the address line instead of binary counter. -- Friendly Regards Tjaart van der Walt tjaartEraseME.....wasp.co.za _____________________________________________________________ | WASP International http://www.wasp.co.za/~tjaart/index.html | | R&D Engineer : GSM peripheral services development | | Vehicle tracking | Telemetry systems | GSM data transfer | | Voice : +27-(0)11-622-8686 | Fax : +27-(0)11-622-8973 | | WGS-84 : 26¡10.52'S 28¡06.19'E | |_____________________________________________________________|

Tom Handley wrote: > I've had a DSO project on the `back burner' for a long time. > I recently had a requirement to analyze IR remote signals so I'm going to finish the project. IR carriers are only around 33-38kHz. IrDA is only 115 kBaud - 4 MBaud. If you have some kind of ADC driven by your printer port, you could still have a very useful gadget with a HUGE memory. Which you can program in C instead of PIC assembly. > Both connect to a bidirectional PC parallel port. > My design will be PIC-based and connect to the serial port. I'd stick with parallel if I were you. You will appreciate it when you want to upload data into your PC. And make sure you design for a commonly available cable such as the Laplink transfer cable. This is only 4 data bits + 1 or 2 handshake bits in either direction. It suits all printer ports, whereas an 8-bit interface might exclude itself from simpler PCs. Either that or try using an LPT<->I2C interface. This won't be slowed by baudrates, won't need a MAX232 chip, or the hassle of baud rate generation. The PIC UARTs don't look very accurate at high speeds, while I2C doesn't need precise clocks. I hear the very fastest scopes use a bit of co-ax cable as a delay line, and digitise signals from points along it. but that's stretching a budget.

part 0 1687 bytes ---------- From: Sean Breheny Sent: Thursday, December 11, 1997 10:36 PM To: EraseMEPICLISTMITVMA.MIT.EDU Subject: Re: Very Fast RAM Well, my idea is somewhere in the neighborhood of 100MHz analog BW. I realize that I don't have that much design experience to get good results up to several 100s of MHz., but I don't see why 100MHz would be too hard. There are lots of video amps and 100MHz op amps out there which are not too expensive and (so it would seem to me) have BW around 100MHz. I have found 30MS/s ADCs for $10. I hope, and I don't see why it would be too hard, to buffer the input of this ADC with a 100MHz op amp, a (roughly) purely resistive attenuator (made from a rotary switch and several resistors), send the output of the ADC to some SRAM, clock the ADC with a PLL with programmable dividing radio (to get adjustable sample rate with very accurate time base), read the output either to a PC or to a microcontroller (PIC :-) ) at a slower rate to feed to an LCD. I could use the multiple sampling technique to digitize at more than 30MS/s as well as go flat out at 30MS/s for up to 15MHz non-periodic stuff. I realize that this would not be as good as a $1000 Fluke scopemeter or similar device, but for my budget, I think it would make an interesting project and be able to visualize waveforms for most hobby/student applications. Am I wrong/wasting my money? Thanks for the help, Sean +--------------------------------+ | Sean Breheny | | Amateur Radio Callsign: KA3YXM | | Electrical Engineering Student | +--------------------------------+ http://www.people.cornell.edu/pages/shb7 RemoveMEshb7EraseMEEraseMEcornell.edu Phone(USA): (607) 253-0315

Keith, thanks for the info. I've done some simple logic analyzers and A/Ds connected to the parallel port with rates around 500Ksps. I've been wanting to do a more versatile, stand-alone, unit for a long time. I wanted RS232 (or 485) so that I can get some distance between the DSO and PC. There are a lot of ways to connect the two and I have'nt decided yet. I have some IrDA eval boards and one possibility is IrDA with fiber or differential drivers instead of IR. Note, this is an in-house project for my own use. I was also looking at flash A/D's and using my Tek 465B as a front-end. - Tom At 02:13 PM 12/12/97 +0000, you wrote: >Tom Handley wrote: >> I've had a DSO project on the `back burner' for a long time. >> I recently had a requirement to analyze IR remote signals so I'm going to > finish the project. > >IR carriers are only around 33-38kHz. IrDA is only 115 kBaud - 4 MBaud. >If you have some kind of ADC driven by your printer port, you could still have a {Quote hidden}> very useful gadget with a HUGE >memory. Which you can program in C instead of PIC assembly. > >> Both connect to a bidirectional PC parallel port. >> My design will be PIC-based and connect to the serial port. > >I'd stick with parallel if I were you. You will appreciate it when you want to > upload data into your PC. >And make sure you design for a commonly available cable such as the Laplink > transfer cable. >This is only 4 data bits + 1 or 2 handshake bits in either direction. >It suits all printer ports, whereas an 8-bit interface might exclude itself from {Quote hidden}> simpler PCs. >Either that or try using an LPT<->I2C interface. >This won't be slowed by baudrates, won't need a MAX232 chip, or the hassle of > baud rate generation. >The PIC UARTs don't look very accurate at high speeds, while I2C doesn't need > precise clocks. > >I hear the very fastest scopes use a bit of co-ax cable as a delay line, and > digitise signals from points along >it. but that's stretching a budget. > >