But the problem with the typical USB logic analysers is that they don't have the elaborate triggering posibilities of a real logic analyser. The ones from Tektronix (and I assume HP/Agilent) have trigger systems that are fully programmable where you can use timers, event counters, states, etc. Basically you can program these using a set of conditions that allow to capture extremely elusive bugs / events. At some point deep memory just runs out.
So the real logic analyzer are Tektronix and HP/Agilent ? For your information an USB logic analyzer can offer much more recording memory and it is quite questionable in most of the cases if timers, event counters, states … are need it at all. But in contrast from Tektronix and HP/Agilent probably those functions can be added.
You have to quantify that... The TLA7AA4 module has 32Mpts per channel. 128Mpts in quarter channel mode. And don't forget timestamped recording which I doubt USB LAs have. I have used the timestamped recording feature on my TLA715 + TLA7AA4 module to capture several frames worth of data from a TFT screen using a 500MHz sampling rate / 2ns time resolution. On top of that the TLA7AA4 has a short memory which is sampled using a sample rate of 8GHz (125ps timing resolution).
Another very usefull feature is to have something similar to segmented recording like you find on oscilloscopes. When a trigger condition is met, the TLA700 series can capture a chunk of samples and then wait for the next trigger. Again, this allows for spanning timeframes that you'll never be able to achieve using a logic analyser that can only do a single shot.
Bottom line: the USB logic analysers are nice for doing protocol decoding and light logic analysis tasks but they are absolutely not a replacement for a real logic analyser. Besides that, real logic analysers can be bought from Ebay for peanuts.
Let's emphasise that with an example...
Consder a system with a 100MHz clock, an address bus, a data bus, and an address valid signal. Every 1ms data is read, and you are only interested in reads from address in the range 1234 to 5678. You want to capture (up to) 10000 values read on the rising edge of the clock provided address valid is asserted and provided the address range is right.
A decent LA will only capture <=10000 values during the 10s window. A low-end LA capture all the irrelevant clock transitions (i.e.2e9 !). It could probably post process those gigasamples to only show all data reads. You would probably have to write your own filter to see only addresses in the range.
That is one very simple example of what a decent LA can do.
Then consider only being interested in those values that occur after the sequence of signal X being asserted followed by signal Y being asserted for more than 1.5ms.
Good luck doing that triggering with a low-end LA!
General point: a good LA allows you to focus on the useful
information, by filtering out the crap from the data.