Yokogawa TA520 Time Interval Analyzer

[capt bullshot]

So I've got this nice instrument for some timenuttery acticities.
The TA520 is a time interval analyzer that can do gapless period width or time delay measurements of signals of more than 10MHz

The unit works nicely as I received it from a seller of used equipment, except when it was operated for more than one or two hours. Then it started to display really strange results:



And the built-in self test reported various failures:



Sometimes it's the Multiplexer Board only, sometimes Multiplexer Board and Aquisition board 2, sometimes Measurement board and Multiplexer board. Turning the unit off, letting it cool down fully restored the normal operation, so this is pointing to some thermally induced failure, like a broken bond wire inside a chip package, bad solder joint, ...

This is a view of the Multiplexer board, I started to diagnose here because this board failed always, the other boards only sometimes.



As there's no service manual of this instrument, the only hint is this block diagram from the user manual:



So, there's four time-to-voltage converters that are driven by the multiplexer and fractional pulse generators. Additionally, in this list view, one can see every fourth measurement is wrong. This gives some hints, one of the four T/V channels might have failed.



I started investigating the multiplexer board, there's just some "standard" ECL chips, none of them ran hot nor was susceptible to thermal shocks (I've used freezer spray on them).
Then I've checked the analog output of the T/V converters and found them all operating. After some more thermal testing (heating or cooling the multiplexer and measurement board and watching the results) I've ruled out measurement and multiplexer as the cause of this failure. There was no obvious correlation between cooling or heating these boards and the occurrence of the failure.

Moving on to the next board in the signal chain, this is the acquisition board 2. It has a set of four large chips, most probably multiplexing and storing the ADC results into the acquisition RAM. The RAM is a bunch of four static RAMs near each large chip. I've started measuring and comparing the RAM signals (data, address, control signals) - and there it is: The RAM chips have the CE2 input unused, and the board designer tied this pin to 5V through a 10k pull-up resistor, one resistor for each group of four RAMs.




Three of the four groups of RAMs had the CE2 pin at high, with some capacitively coupled noise:



At the fourth group, the signal looked different:



Here's the signal at the adjacent pin, A15.



Note the similarity of the waveform, though the amplitude is different. Measuring with a DMM, I found there's a resistance from16k (cold) to 12k (warm) from Pin CE2 to Pin A15 on this group of RAM chips. Now it's quite visible what is the cause for the thermally dependent failure mode: When the instrument is turned on in cold state, the 10k pull-up is enough to keep the level at CE2 just above the "high" input level, and the RAMs are operating. When the unit slowly warms up, this resistance falls to about 12k, the signal level reaches "low" level territory and the RAMs start to act randomly, sometimes storing the acquired values, sometimes not. A perfect match to the symptoms I've observed. A quick test by shorting the pull-up confirmed this.

[capt bullshot]

Now it's about time to locate the offender: Desoldering the RAM chips until the offender is found:



I was lucky at the second chip, so now the root cause was identified.



As I don't have these RAMs in stock, I put the chip back in place for now, and added a 1k resistor in parallel to the 10k pullup. That's sufficient to keep the level at the CE2 pin at "high", so the unit can be used further until a proper replacement is found and ordered.

It's an interesting mode of failure, there must be a resistive path between these two adjacent pins inside the packaging. No external contamination was found.

[Cubdriver]

Excellent troubleshooting, and nice find!  Congrats on figuring it out, and nice that for a change it was the second chip you desoldered that was the problem rather than the fourth, as one would expect to be the case in most instances.

-Pat

[maxwell3e10]

Thanks for sharing your successful repair. I've been curious about Yokogawa time interval analyzers, so looking forward to your tests. In particular, the specs in the datasheet for time jitter is 100 ps rms. For comparison, HP53310 has a spec of 200 ps rms, but in practice gets about 70 ps rms. 

[capt bullshot]

Best results I've seen are in the range from 200ps ... 300ps peak-to-peak and 40ps ... 50ps rms.

[maxwell3e10]

Thanks. So its a bit better than HP53310. I wonder how good it is at frequency measurements. From the manual it seems there is no build-in enhanced frequency measurement. So one would have to save the time stamps for a second or so and fit them to a straight line.

[capt bullshot]

No, it doesn't do frequency measurements at all. And it cannot measure time for more than one period of the input signal on its own (no gate time setting as frequency counters have).
It shows average period / delay time, peak-to-peak and rms jitter (and some other statistics), so one can e.g. measure a 10MHz signal in histogram mode for 1sec, resulting in a histogram of 10 000 000 period times and read the average period time to calculate the frequency. It's not that useful as an frequency counter, nor it can do any enhanced frequency measurements on its own.

[maxwell3e10]

If you can export all time stamps, then one could calculate the frequency. It is a slow process  because it needs to transfer a lot of data to the computer. But in principle it gives more information, since one can calculate the frequency stability for all shorter times as well, by fitting shorter segments to straight lines.

Does it have a setting to measure every Nth edge? That could be used to make a reciprocal counter (if N=10^7 its equivalent to a gate time of 1 sec) or could be used extend the memory to cover a time interval of 1 sec, if N=20.

[capt bullshot]

FYI: I've made some measurements with it and posted the result at the TEA thread:

https://www.eevblog.com/forum/testgear/test-equipment-anonymous-(tea)-group-therapy-thread/msg3712639/?topicseen#msg3712639

https://www.eevblog.com/forum/testgear/test-equipment-anonymous-(tea)-group-therapy-thread/msg3704845/?topicseen#msg3704845

https://www.eevblog.com/forum/testgear/test-equipment-anonymous-(tea)-group-therapy-thread/msg3698053/?topicseen#msg3698053

https://www.eevblog.com/forum/testgear/test-equipment-anonymous-(tea)-group-therapy-thread/msg3651637/?topicseen#msg3651637

(Otherwise, these posts would be difficult to find here)

[capt bullshot]

Of course, one can read the samples and do the postprocessing.
I've got the Programming Manual for the IEEE488 interface, it speaks SCPI and one can read the samples (up to 500000). Haven't done that yet.

No, you can't set it to measure every Nth edge. Either every edge (up to the maximum rate), or you can use the arming function that arms the period counter in regular time intervals, e.g. to measure one period of a 10MHz signal every 1ms (that'll give you 1000 measurements per second, but each measurement is just one period - 100ns). You'd have to add an external divider if you want to measure every Nth edge.

[maxwell3e10]

Thanks.

No, you can't set it to measure every Nth edge. Either every edge (up to the maximum rate), or you can use the arming function that arms the period counter in regular time intervals, e.g. to measure one period of a 10MHz signal every 1ms (that'll give you 1000 measurements per second, but each measurement is just one period - 100ns). You'd have to add an external divider if you want to measure every Nth edge.

This seems annoying. Maybe one could use the inhibit input to prevent it from seeing intermediate edges, but that still requires additional equipment.

Btw, there is also a recent thread on similar HP E1740 https://www.eevblog.com/forum/testgear/hp-e1740a-time-interval-analyzer-292384/  Maybe we need a dedicated thread for time interval analyzers. I've done a bunch of measurements on HP 53310 and Keysight 53230.

[jonpaul]

Bonjour a tous, just seeing the TA320 thread now.

This is in a series of TA320/TA520/TA720 with increasing time res and capabilities, a unique set of devices specifically designed (in 1990s) for jitter measurements eg on CD and disc players and digital audio clocks, etc.

It is much easier to use and more accurate in the histograms etc than the HP5372A etc.

The  frequency function by inverse of period.

The statistics eg sigma are useful as sigma is the wideband RMS jitter.

With low jitter high accuracy 10 MHz   ext clock we can see 25-35 pS.

CH A and CH B will have different min  jitter.

RE experience: We used TA320 and later the much improved TA720 in low jitter audio research in the 1994..2012 period to test and improve AES/EBU, SP/DIF and AES3id digital audio interfaces with FS up to 192..384 kHz;  half bit clk to 24.586 MHz.


RE repairs: A compact puzzle to disassemble lots of very delicate flat cable ribbons to PCB connections.

NO serv manuals available, what existed were in Japanese and in the main plant.

The LCD can get weak, there is a contrast setting in the CFG.

The  internal 3V Li  cell backup,  eventually fails and the unit will not operate. Very deep on the main board. Hours of fun to disassemble and repair and reassemble.

Otherwise no experience in repair, as all four still work fine.


Recommendation: Those  seriously interested in jitter measurement should look for the TA720 they occasionally appear in usual places.

TA320 is much more common and perfectly useable although at lower time resolution.

Bon chance,

Jon

PS: We wrote   three  AES and a  2014 SMPTE paper  on digital audio jitter.

AES 2001: Characterizing Digital Audio Transformers with Induced Jitter Histograms