INSTEK spectrum analyzer GSP-827 -6 dBm down than the real value

[Fraser]

Looking at the pictures of the MMIC gain block stage, there is evidence of rework in the MMIC supply rail and associated bias components (R67,R69 & R71) The MMIC may not be running in optimum conditions and suffering as a result.

As a side note.... I do not know whether the rework was carried out by a service centre, but if so, I would send the tech for retraining ! A pretty poor soldering job on the reworked components. The MMIC installation is very untidy and still has flux residue present. Not good tradecraft at microwave frequencies. Worth removing that flux with IPA even if the stage is working. I would also tidy up the soldering but that may not really be necessary if the stage is operating correctly.

Additional comment: I raise the possibility that R67, 69 & 71 are selected on test in order to 'calibrate' the gain stage and this may be why the bias resistors have been reworked.

Useful app note here:

http://www.avagotech.com/docs/5967-5924E

http://www.hittite.com/content/documents/mmic_amplifier_biasing_procedure.pdf

http://www.ece.ucsb.edu/~long/ece145a/Lab2_145A/BiasingMMICAmps.pdf

Aurora

[videobruce]

Isn't -3db half the level

-3dB is half the power, but if you reduce voltage by a factor of two then power is quartered which is a 6dB drop

Not to get too far OT or hijack the thread, but since you brought it up;
Most SA's that I have seen seem to have a default max 50vdc input level, But the RF levels are different between models with the same max 50vdc. Why?

[Fraser]

An RF spectrum analyser is an ac device and uses capacitor coupling at the input.

To answer your question .... DC blocking capacitor voltage rating versus the sustained ac signal overload capability of the input protection.

Aurora

[G4JMO]

Thanks Aurora for your input. Yes, the MMIC re-work does look untidy work for an official re-work doesn't it!..and the biasing does look like select on test from the resistor chain appearance.
So, I spent the morning checking out the gain from the input through to the output of the MMIC.
Accounting for external cable losses etc, the test setup had 0dBm at 2GHz injected into the input N connector. I then removed the interconnecting coupling caps to measure the signal down the chain.
With C101, 103 removed and coax tapped onto the output of the Pi attenuator, I measured -19dBm. Looking at the Pi components, this seems high..and coincidentally 6dB more than I calculate it should be. But I'll be upsetting the match somewhat with my soldered coax tap and the removal of the coupling caps. I tried to keep the coax tap as short and tidy as possible.
I put the caps back in and then took out C102,104 and tapped onto the output of the switch, U26. I measured -22dBm. This suggests a loss of about 3dB through the six switches. At 0.5dBm per switch, this seems about what I'd expect.
I put the caps back in and then removed the two caps from the output of the MMIC and tapped onto this output. I measured -10dBm, suggesting the MMIC is providing 12dB gain. This is about 3dB lower than the data sheet suggests, but again, I'll be affecting the matching with my coax tap and removal of the coupling caps.

So here's the big surprise. I refitted all the components, and decided to re-check the error on both the 2GHz signal and the 100MHz internal cal signal....
Weirdly, the 2GHz signal was now showing only about 0.9dBm error (reading about -1.9dBm) taking into account the external cable loss of about 1dB.
I switched on the internal 100MHz calibration signal and measured 1dB error.
With all the screening covers screwed back on, the unit is still showing only about 0.9dB error, which is within the tolerances of my test kit and the unit.
Very odd... but I seem to have a correctly working unit after all of that! At least I learnt something about my analyzer   ...but not why it had a 6dB error before stripping down!   

[Fraser]

Interesting news.

A possible interstage coupling capacitor dry joint ?  The 6dB error in the pad is interesting. In such a situation I would have lowered the test frequency to one less affected by cable mismatch etc. 2GHz is one heck of a tricky frequency to carry out a path loss test. I would have plumped for 100MHz or 50 MHz myself as you are looking for an error that is evident at the calibration output running at 100MHz  Far less demanding of your coaxial cables and test equipment 

I hope she keeps working Ok for you

Best Wishes

Fraser

[G4JMO]

Thanks Fraser. Yes, you're absolutely right, 2GHz is making things hard for myself. The reason for the choice of frequency is that I'm without a sig gen that goes below this at the moment. My Giga-tronics 6100 only goes down to 2GHz. Anyway, all seems good. I guess you're right, it could have been a poor interstage  capacitor joint...who knows!

Regards
Andrew

[G4JMO]

...having  warmed up for an hour, the calibration signal is now showing 0.6dB error 

[G4JMO]

A further development!

I left the unit running for several hours. On checking the -30dBm 100MHz internal calibration level, it now showed -36dBm !!   
I tuned the unit off and immediately back on, switched the cal signal on and.......-30.6dBm!
I should have checked the external 2GHz signal too but I switched it off too quickly!
I'll leave it running again and see if it changes and if it does, check internal and external sources.

[Fraser]

Please advise what equipment you are using to RF measure levels.

Grateful if you will confirm that you have tested the level of the calibration output port and ensured that it is stable in amplitude.

Maybe a quick restart from scratch.

Choose a confirmed accurate output signal source. Test the spectrum analyser by viewing the RF source and noting the measured level on different attenuator and IF filter settings. Leave the unit to warm up over several hours and repeat tests exactly as before (including the level of RF source). Compare notes. If an offset is present, switch off SA, restart it and carry out tests again. Note readings Now switch off SA and leave it to cool down for several hours. Switch on and repeat tests.

Using such a test protocol will give us a firm baseline against which to offer advice.

As a side note, most SA's carry out a power on self test diagnostic that also checks the signal path and levels in the IF chain. Not sure whether you unit does this ?

Aurora

[G4JMO]

There is no calibration output port to be able to externally verify the unit's calibration signal level.
External source from my Gigatronics 6100 is being validated on an HP8565A spectrum analyzer.
Since my last post 40 mins ago and restarting the unit, the internal calibration is unchanged at -30.5dBm. The unit has been on continuously for at least 3 hours now, apart from the few seconds to restart it.
 

I've just realised in my previous tests I could have stolen the 100MHz calibration signal from my 8565 instead of using the 2GHz source. Oh well!

If I don't see the reduction again in the next 2 hours, I'll let it cool off for a few hours and then restart from cold and log the amplitude of an external cal signal from my 8565A, which I know is stable, removing one of the variables.

[G4JMO]

Latest test...after the analyzer had been on for 3 hours but briefly powered down to 'reset' the internal cal reading to -30.5dBm as previously described.
Left untouched with the internal cal level being displayed:
(Analyzer settings 10MHz span, RBW 300KHz, VBW, 300KHz Ref level -20dBm)
+1 hr : -30.3dBm
+2 hr : -30.2dBm
+3hr  : -33.9dBm
+4hr  : -34.3dBm
+5hr  : -35.9dBm
+6hr  : -34.0dBm

So a -6dB drift occurred again. but error seemingly reversing and coming down by 2 after the 6th hour.
Disabled internal cal and connected HP8565A cal output to the analyzer. Ext 100MHz reading -15dBm (should be ~ -10 from the HP) indicated a similar ~5dB error to the internal cal source.

Turned the unit off and straight back on and re measured the internal and external cal sources. Internal now back to -30.5dBm and the external reading -11dBm (had to change the ref level to 0dbm to be able to measure the external signal. Changing the ref level from -20 to 0 seems to contribute perhaps 1.5dB of variation)

So this suggests that the calibration amplitude isn't drifting but the analyzer amplitude measurement accuracy is drifting when left running for 3 or 4 hours. But a power recycle corrects the error immediately. Have to assume that some kind of power-up re calibration sequence corrects some parameter for it to regain its accuracy. Perhaps on power-up, the internal cal level is used to characterise and compensate for the gain of a stage that drifts over time?

It will be interesting to hear from Adal as to how his unit varies over time and whether it is similar to  mine and recovers accuracy if power cycled when warm.

[Adal]

Hi Andrew
Thanks for your reply and test. Let me do my own test in order to compare result. I'm out and return the next week to my home lab.
Best regards.

[Fraser]

If this SA was on my bench I would proceed to tap the IF's at various points in the signal path and monitor them as the level error appears. It would then be obvious at which point the error is introduced. I would then investigate the suspect IF stage and look at all signals entering it, including the Local Oscillator level. If all inputs remain constant, yet the stage output drifts, I would start checking individual parts of the stage for the cause.

If the IF stages are fine right up to the final stage I would suspect a Detector/ADC issue, or the processing that follows it.

I treat SA's little different to a Superhet receiver as that is, after all, what they are in terms of RF stages.

I regret that you cannot really diagnose a fault in the IF chain without tapping it at appropriate points. The effect on the IF levels by introducing the coax taps is not an issue as you are interested in Delta IF level and not normal operational levels per-se. Some SA's have IF level monitoring and will throw up an error if the levels are not correct at boot. Really good SA's have diagnostic monitoring capabilities that permit closed case level measurement to spot problems in the RF stages. Sadly I would not expect the GW Instek to have such a capability.

Also, as you suggest, some SA's (like my Advantest 3132) have dynamic calibration that operates at boot, adjusting levels against the accurate calibration signal to provide the correct gain at various points in the IF. A drifting stage will be compensated within a certain range at boot but will continue to drift afterwards as the gain is not retested unless rebooted.

Gain drift could be due to a failure in a gain control circuit that drives a gain variable amplifier stage or there could be an AGC circuit issue on one of the local oscillators. There remains the unpleasant possibility of a a fault in a local oscillator causing level drift at its output.

Kind Regards

Fraser

P.S.

I suggest you do IF signal level monitoring with the SA centred on a known RF test source and the span set to 'Zero Span'.

[Fraser]

Just had another random thought..... if any of the local oscillators are drifting in frequency you could be seeing the IF signal drifting into the skirts of the IF filter and so reducing the level passed to the next stage. Just a thought.

Aurora

[G4JMO]

Thanks for your thoughts Fraser. All very sensible ideas.

Unlike my old HP8565A analyser which is straightforward to work on (helps having a service manual too!), the Instek is of the 'Crunch n FFT' variety. Given the construction (photo attached) it's not that easy to work on either.

Without a service manual and beyond the accessible front end RF stages, it's a bit of a Black Hole. Being pragmatic,  given the nature of the problem, I can work around it once I've  characterised the  drift, and it proves to be predictable.

As long as it has warmed up, a quick power cycle seems to keep it within a dB for at least 2 or 3 hours. I can put up with that! 

I've just started a new test from cold tonight. It started out at -34.5 dBm and after an hour  it's sitting at -30.9 dBm. We'll see if it drifts up again.

Best regards
Andrew

[G4JMO]

Results from the 'start from cold' test...

The same drift profile occurred. After settling at ~ -31dBm the level stayed more or less constant for about 3 hours. Then in the space of about 1 hour, the level drifted to ~-35dBm. Power cycling returned the reading to -31dBm. So as long as I remember to power cycle after its normal stabilisation period, it's manageable.

The question remaining is whether Adal see's the same kind of drift profile and recovery or whether his unit is suffering from a different problem.

[Adal]

Hi everybody.
Here are the result of my own test
After 6 hours my GSP-827 Shows -35 dBm in the calibration signal. No changes if I cycle on-off the spectrum analyzer.

[KJDS]

Hi everybody.
Here are the result of my own test
After 6 hours my GSP-827 Shows -35 dBm in the calibration signal. No changes if I cycle on-off the spectrum analyzer.

That's an odd looking shape. Can you halve the bandwidth but leave the RBW the same, then check the level with each of the RBWs, you'll probably need to reduce the span for the narrow bandwidths.

[Adal]

KJDS here are the result of one test. No matters what RBW use, the error level remains the same. I need check the input path and check again

[Adal]

Here are the result using the same settings of Andrew spectrum analyzer. ?

[KJDS]

When I saw the first plot I thought it possible that one of the secondary LOs had become unlocked, causing the IF to miss the filter, hence the odd shape, but a wider RBW would have shown lower loss and that's not happening.

Unless you can probe down the IF path then you won't work out where the issue is.

[Adal]

Ok KJDS.  Thanks for your comment.
Andrew, could you explain me how take apart the covers? I can't take over the covers?

[Adal]

All the screw are out, but the cover does not slip. I can't identify the cause. Maybe Andrew can help me

[G4JMO]

Hi Adal,
I was initially puzzled by the cover too! ...then I discovered that the carrying handle end caps push off and reveal two more screws. Undo these and then the cover slips off.
Best regards
Andrew

[Adal]

Here are the inner's of my spectrum analyzer. The input path and the components that Andrew check first. I will testing tomorrow and let you know