£150-ish logic analyser

[Mario87]

Hi all,

I'm looking to get a logic analyser, specifically for inspecting the RGBA output of a GPU. This is a 32-bit data stream, so 32 channels or more would be ideal and the GPU clock frequency is around 147 MHz, so I assume I'll need an analyser which can support that kind of bandwidth.

I have been looking at the Hantek 4032L but have read the claimed 150 MHz input is maybe not so accurate. See posts on page 2 of this thread.... https://www.eevblog.com/forum/testgear/usb-logic-analyzer-whats-the-current-favorite-for-~150$/25/

I have also been looking as the DSLogic Plus, which is only 16-channel, but I was wondering, is it possible to purchase 2 of them and have both sampling data simultaneously? Will the DS View software allow that??

My main concern with the DS Logic is they officially state only 100 MSPS when sampling all 16 channels, but then they also state a bandwidth of just 50MHz, where as in post 22 of the link above it seems the DS Logic can actually handle around 200 MHz at the input?

Thoughts & suggestions welcome. Please don't say "go and buy a £500 analyser" though. 

[rsjsouza]

Thoughts & suggestions welcome. Please don't say "go and buy a £500 analyser" though. 

Yeah, right... 

The DSLogic U3Pro32 seems to fit the bill. However, their software hasn't seen an update in about a year now - although the lack of updates is not necessarily a bad thing, it might indicate lack of a roadmap.

The DSLogic Plus has a TI (Trigger Input?) signal that might allow to do what you want, but the documentation says it is "Reserved". The software itself does not control multiple probes at the same time - perhaps run two instances?

I have a Kingst VIS LA2016 and its software is quite nice; their MIPI16 model has a compatible sample rate, but also only 16 channels.

[Fungus]

Digital Discovery:

https://store.digilentinc.com/digital-discovery-portable-usb-logic-analyzer-and-digital-pattern-generator/


Edit: Oh, wait... it's "only" 200MS/s with all 32 channels enabled.

[Mario87]

Digital Discovery:

https://store.digilentinc.com/digital-discovery-portable-usb-logic-analyzer-and-digital-pattern-generator/


Edit: Oh, wait... it's "only" 200MS/s with all 32 channels enabled.

Oh! This one could be interesting @ £170 from Farnell UK. I know its 200MS/s with all 32 channels enabled, but I assume that's 200MS/s per input? So if I am monitoring 32 inputs running at a frequency of 147 MHz then sampling at 200 MHz on each input should still ensure I do not miss any signal edges.

Or am I missing something here (been one of those days, so probably am missing something obvious)?

[MarkL]

How much sample memory do you need?  Do you need timing and/or state mode?

A lot of the old boat-anchor logic analyzers have at least 32 channels, and 150MHz data rate for state mode and (much) higher for timing mode is not a problem.  HP/Agilent 16xx/16xxx series and Tektronix TLA series come to mind.

[Fungus]

Digital Discovery:

I know its 200MS/s with all 32 channels enabled, but I assume that's 200MS/s per input?

Yes.

So if I am monitoring 32 inputs running at a frequency of 147 MHz then sampling at 200 MHz on each input should still ensure I do not miss any signal edges.

Or am I missing something here (been one of those days, so probably am missing something obvious)?

Doesn't "147MHz" mean 294 million signal changes per second?

[Pinkus]

Intronix Logicport?

[0xdeadbeef]

Please don't say "go and buy a £500 analyser" though. 

The DSLogic U3Pro32 seems to fit the bill.

The DSLogic U3Pro32 is currently sold for  $399 (witout taxes and shipping). So that's about twice of what the OP wanted to pay.
Anyway, IMHO the the "normal" DSLogic PLus is still the best bang for the buck. Yes, the SW isn't perfect, and the triggering is somewhat limited, but other than that, it's a pretty nice LA.

About bandwidth: no normal LA has a terrific analog bandwidth. That's kinda impossible with the plain cables they use as probes. At least the DSLogic has shiedled probes which help to avoid inductive loops.
Anyway, a LA only makes sense if the shape of the signal is OK, so the LA doesn't need to have unlimited bandwidth just to detect the correct levels. At some point, too little bandwidth might shorten measured times due to the edges becoming too slow/soft.  Indeed, measuring a ~150MHz digital signal will be problematic with any normal LA, no matter if the claim to have 200MSa/s sampling rate or 500MSa/s. Then again, usually the CPU clock is not necessarily a good indication for the maximum frequency you need to measure. It seems somewhat unlikely that a 150MHz CPU can create a 150MHz digital signal at its output. Even if it would be able to alter the pin level at every clock cycle, the maximum output frequency would we 75MHz.

[Fungus]

It seems somewhat unlikely that a 150MHz CPU can create a 150MHz digital signal at its output. Even if it would be able to alter the pin level at every clock cycle, the maximum output frequency would we 75MHz.

True.

Some chips (eg. Atmel Tiny85) have an internal PLL that can drive their internal timers at higher speeds than the CPU base frequency but that's unlikely to apply to a data bus. A data bus usually runs at half the CPU clock (or less...)

[Mario87]

How much sample memory do you need?  Do you need timing and/or state mode?

A lot of the old boat-anchor logic analyzers have at least 32 channels, and 150MHz data rate for state mode and (much) higher for timing mode is not a problem.  HP/Agilent 16xx/16xxx series and Tektronix TLA series come to mind.

Honestly not sure about how much sample memory I would need, probably enough to capture a few seconds of data. One reason I was looking at USB analysers is other than sample memory for buffering the amount of data which can be captured is rather unlimited.

Would mostly be used in timing mode, but state mode could prove useful for some projects.

I know its 200MS/s with all 32 channels enabled, but I assume that
Doesn't "147MHz" mean 294 million signal changes per second?

Haha, I knew i was overlooking the obvious! Yes, 1 clock cycle will be the signal going positive and negative, so 2 state changes which means 147 MHz clock will be 294 million state changes per second.

Intronix Logicport?

Looks great, but it’s €425 + VAT & duty, etc (thanks Brexit) as I can’t see a UK seller for it, would make it well over €500 delivered.

The DSLogic U3Pro32 is currently sold for  $399 (witout taxes and shipping). So that's about twice of what the OP wanted to pay.
Anyway, IMHO the the "normal" DSLogic PLus is still the best bang for the buck. Yes, the SW isn't perfect, and the triggering is somewhat limited, but other than that, it's a pretty nice LA.

About bandwidth: no normal LA has a terrific analog bandwidth. That's kinda impossible with the plain cables they use as probes. At least the DSLogic has shiedled probes which help to avoid inductive loops.
Anyway, a LA only makes sense if the shape of the signal is OK, so the LA doesn't need to have unlimited bandwidth just to detect the correct levels. At some point, too little bandwidth might shorten measured times due to the edges becoming too slow/soft.  Indeed, measuring a ~150MHz digital signal will be problematic with any normal LA, no matter if the claim to have 200MSa/s sampling rate or 500MSa/s. Then again, usually the CPU clock is not necessarily a good indication for the maximum frequency you need to measure. It seems somewhat unlikely that a 150MHz CPU can create a 150MHz digital signal at its output. Even if it would be able to alter the pin level at every clock cycle, the maximum output frequency would we 75MHz.

Hmm, interesting information, so you seem to think any of the cheaper ones would work (Digilent Digital Discovery, DSLogic Plus or Hantek 4032L)?

Might try and measure 1 or 2 lines with the oscilloscope to see what frequency I am actually dealing with on these specific pins I’m looking to measure. Then I suppose I can make a more informed decision.

Only thing that pushes me away from the DSLogic is the channel count. Really need more than 16 channels. However the Digital Discovery is really tempting if the actual frequency on those pins is under 100 MHz with its 200 MS/s sample rate, should do the trick.

True.

Some chips (eg. Atmel Tiny85) have an internal PLL that can drive their internal timers at higher speeds than the CPU base frequency but that's unlikely to apply to a data bus. A data bus usually runs at half the CPU clock (or less...)

Yeah, think I need to check with the scope what frequency those pins are operating at and make a decision based on that,  it sounds like the Digital Discovery will be the one to suit my needs & price point.

[rsjsouza]

Please don't say "go and buy a £500 analyser" though. 

The DSLogic U3Pro32 seems to fit the bill.

The DSLogic U3Pro32 is currently sold for  $399 (witout taxes and shipping). So that's about twice of what the OP wanted to pay.

You edited out my previous remark.

[Doctorandus_P]

As a rule of thumb you want your Logic analyser to sample at least at 4 times your signal frequency. Maybe more if you want to verify setup times and other relative timing between signals.

Signal integrity during sampling is also becoming increasingly difficult at higher frequencies.

And about the number of channels... Do you want 32 or do you need more?

From the little info you've given I'm guessing that USD500 will get you closer to the performance you want.

[Fungus]

Please don't say "go and buy a £500 analyser" though. 

I'm guessing that USD500 will get you closer to the performance you want.

What's the exchange rate today? 

[Electro Fan]

https://www.xe.com/currencyconverter/convert/?Amount=150&From=GBP&To=USD

[Mario87]

As a rule of thumb you want your Logic analyser to sample at least at 4 times your signal frequency. Maybe more if you want to verify setup times and other relative timing between signals.

Signal integrity during sampling is also becoming increasingly difficult at higher frequencies.

And about the number of channels... Do you want 32 or do you need more?

From the little info you've given I'm guessing that USD500 will get you closer to the performance you want.

Thanks for the info. What issues would there be with using say a 200MS/s LA with a 75MHz signal? Why is the rule of thumb 4x the frequency? I understand with a scope you need a bandwidth at least 5 times that of the signal frequency so that you pick up the 5th harmonic on a square wave instead of just a single harmonic (sine wave). Not sure why 4x for a LA though??

I will try and determine the actual the frequency of the signals with a scope this week. This is just a side project, so just as & when I get time to look at it I will.

As for number of channels, I need at least a 32 channel LA, but if I can get one with more that also fits the bill, then I would certainly consider it.

$500 is still £360 though 

[egonotto]

Thoughts & suggestions welcome. Please don't say "go and buy a £500 analyser" though. 

Hello,

sorry I doubt you find a LA for £500 that can do what you want.

I believe you need a sample rate of 1GS/s. This is 4GB/s. If you want see 5s you have 20GB.

Even if 0.5GS/s is sufficient it helps not much.

Best regards
egonotto

[Mario87]

Thoughts & suggestions welcome. Please don't say "go and buy a £500 analyser" though. 

Hello,

sorry I doubt you find a LA for £500 that can do what you want.

I believe you need a sample rate of 1GS/s. This is 4GB/s. If you want see 5s you have 20GB.

Even if 0.5GS/s is sufficient it helps not much.

Best regards
egonotto

Can you elaborate where you get these numbers from as they don't seem to make sense or tie in with what others have said? Also, 1GS/s would be 125MB/s (8 bits in a byte) which for 5s would be 625MB of data.

Regardless though, the clock speed of the GPU is 147 MHz, so if you are implying I need to sample at 1GS/s that would be about 6-7 times the internal clock speed and possibly 12-14 times the bus speed (still need to confirm what the bus speed is), which seems a bit overkill.

[Fungus]

1GS/s would be 125MB/s (8 bits in a byte) which for 5s would be 625MB of data

Per channel.

You want 32 channels so it's 32x that much.

[Fungus]

Regardless though, the clock speed of the GPU is 147 MHz, so if you are implying I need to sample at 1GS/s that would be about 6-7 times the internal clock speed and possibly 12-14 times the bus speed (still need to confirm what the bus speed is), which seems a bit overkill.

You can never have too much. 

Relying on mathematical theory is one thing, real life is another. In practice there's a couple of problems:
a) All the cables and connectors and traces on the PCB will reduce the bandwidth by tiny amounts so aiming to run right on the limits is a bad idea.
b) When you start approach the limits then you have to take into account the phase of your sampler vs. the phase of the clock of the thing you're looking at. If they're nicely in phase then it all works beautifully, if they're out pf phase then it all goes to hell. Murphy's law says they'll always be out of phase so it's best to have some margin of oversampling to avoid this. Maybe 2-3x, so 500MHz wouldn't be "overkill" for looking at a 150MHz clock.

In short: You're asking a lot for your 150 quid.

[tggzzz]

Why is the rule of thumb 4x the frequency?

Not necessary where the LA's clock is derived from the UUT's clock - in that case you only need to sample once per bit period. More usually that isn't possible and you have to use asynchronous sampling, in which case to guarantee to capture each bit you need at least two samples per bit period.

I understand with a scope you need a bandwidth at least 5 times that of the signal frequency so that you pick up the 5th harmonic on a square wave instead of just a single harmonic (sine wave).

No, you don't understand; the only thing that is important is the risetime. For a little theory and some measurements, see https://entertaininghacks.wordpress.com/2018/05/08/digital-signal-integrity-and-bandwidth-signals-risetime-is-important-period-is-irrelevant/

[Mario87]

Per channel.

You want 32 channels so it's 32x that much.

Yes, very true, with 32 channels that would be 20GB of data in 5s.

You can never have too much. 

Relying on mathematical theory is one thing, real life is another. In practice there's a couple of problems:
a) All the cables and connectors and traces on the PCB will reduce the bandwidth by tiny amounts so aiming to run right on the limits is a bad idea.
b) When you start approach the limits then you have to take into account the phase of your sampler vs. the phase of the clock of the thing you're looking at. If they're nicely in phase then it all works beautifully, if they're out pf phase then it all goes to hell. Murphy's law says they'll always be out of phase so it's best to have some margin of oversampling to avoid this. Maybe 2-3x, so 500MHz wouldn't be "overkill" for looking at a 150MHz clock.

In short: You're asking a lot for your 150 quid.

You are 100% right, can never have too much, but also no point in buying the best / most expensive kit if its not needed. On the other hand, also no point in buying cheaper kit if it wont do the job. All about balance I suppose.

As for point b, I have found a schematic for this GPU and it does look like there is an output clock which goes to a DAC (which all the data lines also go to) before being output over A/V to a display. So I may be able to use that clock to sync with the data lines and capture data in state-sampling mode with the LA. Should hopefully mean I can get away with a LA more within my budget.

I don't mind paying a bit more for a LA if it will do the job. £150 is not set in stone, but when going up to £400, £500 or more it's getting tough to justify for this which is really a passion project, not a work related thing.

Not necessary where the LA's clock is derived from the UUT's clock - in that case you only need to sample once per bit period. More usually that isn't possible and you have to use asynchronous sampling, in which case to guarantee to capture each bit you need at least two samples per bit period.

I have found a schematic for this GPU and it does look like there is an output clock which goes to a DAC (which all the data lines also go to) before being output over A/V to a display. So I may be able to use that clock to sync with the data lines and capture data in state-sampling mode with the LA. Should hopefully mean I can get away with a LA more within my budget, but I need to try and confirm this clock is for those data lines and not something else.

No, you don't understand; the only thing that is important is the risetime. For a little theory and some measurements, see https://entertaininghacks.wordpress.com/2018/05/08/digital-signal-integrity-and-bandwidth-signals-risetime-is-important-period-is-irrelevant/

Interesting read, thanks for the link.

[tggzzz]

No, you don't understand; the only thing that is important is the risetime. For a little theory and some measurements, see https://entertaininghacks.wordpress.com/2018/05/08/digital-signal-integrity-and-bandwidth-signals-risetime-is-important-period-is-irrelevant/

Interesting read, thanks for the link.

You're welcome, and check whether the LA really has a clock input. Look for the asynchronous and synchronous specification.

Most modern ones "synthesise" a clock by you defining which is the clock input and still asynchronously sampling - but post-processing the captured signals to look for clock transitions.

[Mario87]

Hi all,

I FINALLY got in the position to measure the buss frequency with the scope. It’s 54Mhz, so that digilent 200MS/s analyser with 32 channels  for £170 should do the job as it’s going to be sampling at almost 4x the signal frequency.

Thanks for everyone’s input! 

[egonotto]

Hello,

@ Mario87: have you further experience with digital discovery with your project?

Best regards
egonotto

[Doctorandus_P]

I've never used a logic analyzer with more capabilities then the EUR10 dongle with an CY7C68013A, which samples at most at 24MHz myself. I would sort of like an upgrade, but it's not important enough for me to invest serious money in though.

Maybe have a look at:
https://sigrok.org/wiki/Supported_hardware#Logic_analyzers

Even if you don't want to use Sigrok / Pulseview, it is a nice overview of logic analyzers.

A project I would find very interesting, is  if someone took the firmware for the CY7C68013A, which is FX2, and made it compatible with the FX3 & Sigrok for higher sampling frequencies and more channels.

I've also read some stories of people smarter then me who do FPGA development, and just program in a LA as part of the program for debugging the FPGA. Such a project would be relatively easy to extend with external inputs.