Looking to buy an oscilloscope for home use, I've seen the questions on here in regards to bandwidth. I'm interested in prototyping microprocessors with clock speeds 800+ MHz, but I shouldn't need an oscilloscope with that bandwidth, correct? Only for the external buses/peripherals, so if the memory bus was 100MHz then I'd need a scope with at least 100MHz bandwidth, so on and so forth. Do I have the right idea??
\$\begingroup\$
\$\endgroup\$
4
-
3\$\begingroup\$ If you're interested in playing with 800 Mhz + MCUs, expect to be investing $50K+ in tools. They're non-trivial. \$\endgroup\$– Connor WolfCommented Dec 26, 2013 at 4:52
-
1\$\begingroup\$ I see, the same tools(scopes, analyzers, etc), only higher bandwidth? I know the cost increases almost exponentially it seems. \$\endgroup\$– MDMoore313Commented Dec 26, 2013 at 5:04
-
2\$\begingroup\$ It's other things. MCUs in that speed-range are all many, many pin BGA, so you're going to need X-Ray inspection facilities for assembly (or to farm it out - EXPENSIVE). Also, you'll need active probes to not load your busses down, and you should probably go with name-brand tools (e.g. Agilent) so you have access to support. \$\endgroup\$– Connor WolfCommented Dec 26, 2013 at 21:45
-
1\$\begingroup\$ Spent a career supporting high speed uProcessor characterization on $$up to $6million dollar automated test equipment. From experience. Speed kills. Some of the stuff I see today for >10Gbit stuff... scopes cost as much as a house, and scope probes each as much as a BMW (used). \$\endgroup\$– Ross YoungbloodCommented Dec 11, 2023 at 16:50
Add a comment
|
3 Answers
\$\begingroup\$
\$\endgroup\$
2
Processor are complex chips regarding their clock : they use a crystal/oscillator as a primary clock source. But this primary clock source is lower than 100MHz, so a PLL is used to multiply and "clean" the clock signal. Then the clock will be used in two domains : core clock and bus/peripheral clock, with the core clock running at the highest frequency.
As far as I seen, the core clock is rarely routed on a pin, so you can't observe it. And, actually, we don't care if it's available externally or not : you don't have access to the internal CPU either.
So what is important is what is going in and out of the chip, not what is inside. What are the bus and communication protocols used ? What are their speed and bandwidth ? Are they serial / parallel / differential ?
- Oscilloscope bandwidth:
As Rolf Ostergaard said : \$BW = \frac{0.35}{t_{rise}}\$
If you don't know the rise time, for instance when there is no minimum rise time in the datasheet, you have two cases :
- Quick look at a signal : in this case at least 5 times the signal frequency is good.
- Detailed observation (rise/fall, measurement) : in this case at 10 times will be necessary (like AngryEE said)
That's what I'm trying to do at work, but most of the time I have 500 MHz / 1 GHz oscilloscope for under 50 MHz signals, so I'm always good
- Oscilloscope sample rate:
Just check that the sample rate is 3 to 4 times the oscilloscope bandwidth (or your signal frequency)
- Probes:
Don't forget the probes ! Check for their bandwidth too. Also as signals higher than 100MHz are frequently differential, you may need a differential probe.
A good feature when looking for glitch is the FastAcq (Tektro), WaveStream (Lecroy) : this is a statistical representation of the waveforms.
Logic Analyzer are not used to observe a waveform but to analyze timings, hence the name. With an oscilloscope you will check the signal integrity of the clock and data signal of your bus, whereas with your logic analyzer you will check the timings.
-
1\$\begingroup\$ Scopes are often sold with an given bandwith, say Rigol DS2202, 200Mhz, but it's sample rate is 2GSa/s in one channel, and 1GSa/s in dualchannel. As noted abowe, is it still the bandwith that need to be like 5-10x or is it the Samplerate of the scope? \$\endgroup\$ Commented Feb 12, 2014 at 18:15
-
1\$\begingroup\$ BW is calculated with the rise time of your signal, so if you want to observe signals with a low rise time, if your oscilloscope have a low bandwidth, you won't be able to see the rise correctly. The Rigol example you're mentionning is useful is you want to capture a "glitch", but you're still limited by the BW... If you observe a 100 Mhz equivalent rise time signal and you suspect a glitch occuring with a 500MHz equivalent rise time, the BW is limiting. But observing 10 periods of a 10MHz equivalent rise time signal, then with a 2GSa/s, you'll be able to see a "fast" glitch. \$\endgroup\$– zeqLCommented Feb 12, 2014 at 19:32
\$\begingroup\$
\$\endgroup\$
6
The bandwidth specifications on an oscilloscope are for sine waves only. For example, if your oscilloscope has 100MHz bandwidth that means that it won't have any problem with 100MHz sine waves. 100MHz square waves are a different matter - generally to get a good look at a square waveform you'll need 10x the bandwidth. So, for a 100MHz square wave to show up nicely on your oscilloscope you'll need 1GHz bandwidth.
-
\$\begingroup\$ Thnx, 1GHz scopes are quite pricey, would a logic analyzer be best for that sort of thing?? \$\endgroup\$ Commented Dec 26, 2013 at 1:37
-
1\$\begingroup\$ I would go with a logic analyzer first. I find one to be more useful (but then, I work mostly on projects at the low-MHz range) than an oscilloscope. \$\endgroup\$– RenanCommented Dec 26, 2013 at 2:03
-
\$\begingroup\$ @Renan Are you saying logic analyzers are cheap? ;o) \$\endgroup\$– jippieCommented Dec 26, 2013 at 10:22
-
2\$\begingroup\$ And to get 1 GHz bandwidth you need a sampling rate at least 2.5GHz and probably higher, if it's a digital scope. (you can pry my Tek2465b out of my cold dead hands!) \$\endgroup\$– user16324Commented Dec 26, 2013 at 11:11
-
1\$\begingroup\$ @MDMoore313 - Logic analysers and Oscilloscopes are different tools. You will probably need both. \$\endgroup\$ Commented Dec 26, 2013 at 21:46
\$\begingroup\$
\$\endgroup\$
1
When buying an oscilloscope for digital stuff, you really should be thinking of what rise/fall times you want to be able to see.
Use the BW (GHz) = 0.35/t_rise (ns) formula to get an idea of the 3dB bandwidth of that edge.
Modern parts like DSP's, memories, FPGA's etc. easily have rise/fall times as fast as 200ps, meaning you need at least 0.35/0.2 = 1.75GHz bandwidth in probe and scope. This explains why 2GHz bandwidth scopes seem to be quite popular in professional setups these days.
This is not cheap, and you can get by with less - just remember you may not be able to fully see the fast edges.
-
\$\begingroup\$ Thanks, exactly what I was looking for. Makes perfect sense actually, rise/fall is what you want to see. \$\endgroup\$ Commented Dec 26, 2013 at 12:55