Lukas,
Interesting links, I not sure if the OP is trying to measure a Half Bridge or Full Bridge AMP design - and what sound card he's using as I ponder if he also needs a Diff to SE conversion as well as a LPF.
The Elektor kit uses ferrite cores, from my experience ferrite core based inductors bottom out at about 0.001% THD - maybe they have found a better core material - they don't mention the use of airgaps to improve linearity.
Interesting links, I not sure if the OP is trying to measure a Half Bridge or Full Bridge AMP design - and what sound card he's using as I ponder if he also needs a Diff to SE conversion as well as a LPF.
The Elektor kit uses ferrite cores, from my experience ferrite core based inductors bottom out at about 0.001% THD - maybe they have found a better core material - they don't mention the use of airgaps to improve linearity.
Hi John,
Yes, I referred to the Audio Precision usage of the AES-17 filter. A simple R-C filter gives about 20 dB attenuation and should be ok for "normal" Class-D amps.
Interesting note you made about the AD797. Did'nt think of the degeneration resistors. AD claim that the AD797 holds its superb noise performance up to very high frequencies due to the single stage design.
Can you post a graph of the R&S THD+N over frequency if you find time? Would be
interesing to see the comparison to AP Sys2.
We have at work a lot of UPV and UPD but none with the low distortion generator option :-/
But i can post a graph of the newest AP 555 machine
Greetings,
Udo
Udo,
If you ever get a chance to visit me here in Czech then we can run through some Opamp RF IMD tests. I have to perform the test manually, based upon a concept developed by Paul miller.
I did Email you awhile back to see if you would be interested in for a fee writing a test procedure for me using the GPIB interface BUS to automate the tests:-
https://dl.dropboxusercontent.com/u/86116171/Resonances and Repercussions.pdf
Here is more on the subject, ignore my mention of the Labview environment as apparently not even LabView's own Mother loves it:-
http://www.diyaudio.com/forums/equipment-tools/274655-anyone-here-have-experance-ni-labview.html
I meant to mention earlier that the fact the the AES17 filter improves a 20KHz BW limited measurement just highlights the RF non linearity of the test equipment.
If the test equipment had perfect behavior then the filter should have zero impact on the measurement results (presuming the same measurement BW).
It sound not be considered that the filter it artificially "Improving the measurement figures" rather its give more accurate measurement results by reducing the non linearity of the test equipment, and not somehow improving the linearity of the UUT.
I'll post A THD+N plot later - yesterday my UPD started puffing smoke from its side vents, a Ceramic Cap popped making a real mess of the PCB - what a nice Birthday present
https://dl.dropboxusercontent.com/u/86116171/IMG_7352.JPG
Ferrites have a reputation for nonlinearity but the Radiometer CLT-1 uses them for both the filters and the matching transformer and it has a residual of less than -165 dB so I think its not that simple. I reverse engineered the AP passive filter some time ago. It also used ferrite inductors as I remember.
Most digital amps seem to bottom out at .01% so it may not be an issue on the other side. If you are working with Bruno's amps (Hypex) you probably are not obsessing about THD, since he did the work for you.
I would argue that a differential input is required to test amps. The return current from the load to the amp ground getting mixed with the test signal from having a common ground for both can at a minimum alter the actual measurement and could lead to all sorts of mischief and instabilities. The one aspect missing from the eBay filter is an attenuator. Of course an attenuator could be part of a load resistor anyway. The LinearX version has an attenuator but the steps would drive me nuts for the scaling.
Most digital amps seem to bottom out at .01% so it may not be an issue on the other side. If you are working with Bruno's amps (Hypex) you probably are not obsessing about THD, since he did the work for you.
I would argue that a differential input is required to test amps. The return current from the load to the amp ground getting mixed with the test signal from having a common ground for both can at a minimum alter the actual measurement and could lead to all sorts of mischief and instabilities. The one aspect missing from the eBay filter is an attenuator. Of course an attenuator could be part of a load resistor anyway. The LinearX version has an attenuator but the steps would drive me nuts for the scaling.
If I understand the Miller test linked, its using a pseudorandom modulation of an RF carrier and plotting the resulting audio spectrum. It should be pretty easy to do. I'm not sure why the modulation needs to be pseudo-random. I would think two RF carriers with a sub 20 KHz difference would accomplish about the same and be easier to detect? Is there something about the RF sensitivity that is greater to random noise? Maybe its very frequency specific? It does not look to be in the plots. A wideband RF noise generator could also be used it seems. High pass filter it at say 100 KHz or 1 MHz and gradually increase the level and plot the noise spectrum on the output of the device.
Its hard to say as each situation is different - but I guess Pauls thinking was to modulate the whole 20KHz baseband "in a single go" - if you look at the Power amplifier plots you can see their sensitivity to RF gets worst towards 20KHz (at certain RF frequencies) - spot modulation frequency's would not clearly represent whats really happening across the Baseband.
The bottom the graph is scaled 20Hz to 20KHz, so a 20KHz BW limited noise source is the perfect stimulus.
Really today it would be better to extend the baseband to say 100KHz.
The bottom the graph is scaled 20Hz to 20KHz, so a 20KHz BW limited noise source is the perfect stimulus.
Really today it would be better to extend the baseband to say 100KHz.
I've a couple of the AUX0025 sitting stacked up on the bench in front of me, I'm pretty sure they are aircores - although I've not looked inside one for 10 years or so...
Thats typical, but does not have to be the case, I've built ClassD amplifiers that are more linear then the AP SYS2.
UCD uses post filter feedback so your not left at the mercy of the inductors linearity.
I'm just looking for a simpler way to test for the same issue. I think the following would be a possibility- Use a sweepable RF generator with AM modulation capability. AM modulate it with 10 KHz at 100% witch will have two spectral sidebands at +/- 10 KHz from the now missing carrier or in effect two carriers 20 KHz apart. Then slowly sweep it from say 200 KHz (if possible) to 200 MHz and plot the output at 20 KHz (demodulated difference IM). This should be a good proxy for RF nonlinearity and spot specific frequencies where problems exist. It would be much easier to sense the specific 20 KHz even in noise than trying to spot pseudorandom noise in random noise. Quicker as well. Is there some aspect this would miss?
Here's the odd thing, lets say when an opamp is subjected to RF, you don't see discrete sidebands or fixed spurie (atleast within the audio Baseband) but a wideband elevated noise floor - it might only be elevated by 2dB to 3dB but its elevated non the less.
A slightly different scenario, but same result is that on numerous occasions I’ve caught a high speed Opamp oscillating due to its elevated baseband noise floor. Typically the noise floor would be elevated by no more then say 10dB, so you need to have an idea where the noise floor should be otherwise you would miss it.
If you look at Paul's graphs you will see that the noise floor is scaled from 0dB to +15dB, this matches my own experience where I would typically see around 10dB noise floor modulation on a bad day…
This is the process I'd like to automate:-
1. Perform FFT of UUT noise floor across 20KHz or 100KHz span (With no injected RF stimulus). Store this noisefloor result as the Reference baseline noise-floor (the UUT will be an amplifier or some form of analogue gain stage etc).
2. Sweep a RF signal generator from say 20KHz to 3GHz, or a user set frequency span. The RF Signal Gen set to 20mV RF carrier level with AM 100% modulation via an external random noise source B/W limited to 20KHz.
3. At frequency points along the RF sweep, perform a Analogue FFT and plot result on a colour graded waterfall showing any noise floor modulation as a result of the external RF carrier injected stimulus (any RF induced noise floor modulation when compared to the earlier stored Reference noise floor without the external RF signal present).
A slightly different scenario, but same result is that on numerous occasions I’ve caught a high speed Opamp oscillating due to its elevated baseband noise floor. Typically the noise floor would be elevated by no more then say 10dB, so you need to have an idea where the noise floor should be otherwise you would miss it.
If you look at Paul's graphs you will see that the noise floor is scaled from 0dB to +15dB, this matches my own experience where I would typically see around 10dB noise floor modulation on a bad day…
This is the process I'd like to automate:-
1. Perform FFT of UUT noise floor across 20KHz or 100KHz span (With no injected RF stimulus). Store this noisefloor result as the Reference baseline noise-floor (the UUT will be an amplifier or some form of analogue gain stage etc).
2. Sweep a RF signal generator from say 20KHz to 3GHz, or a user set frequency span. The RF Signal Gen set to 20mV RF carrier level with AM 100% modulation via an external random noise source B/W limited to 20KHz.
3. At frequency points along the RF sweep, perform a Analogue FFT and plot result on a colour graded waterfall showing any noise floor modulation as a result of the external RF carrier injected stimulus (any RF induced noise floor modulation when compared to the earlier stored Reference noise floor without the external RF signal present).
Last edited:
I follow the noise floor increase and have seen that myself. 3 GHz RF generators are not common in DIY but the basics are available. Maybe it would be possible to link something with an FFT package that can be linked, Excel (which can drive GPIB with the right bit of software) and someone who has done some of this.
100% modulation by noise is a misnomer since noise has a crest factor that can be pretty high. However I understand what you are after. The NI stuff is a major commitment to get into which is why I have not gone near it. Using a soundcard for the FFT would be obvious. Excel? or Sigview spectrum analyzer - FFT based signal analysis software Sigview may be able to do the whole task. It would be worth an e-mail. If so I'll get a copy as well. And then scrounge a GPIB equipped RF generator.
100% modulation by noise is a misnomer since noise has a crest factor that can be pretty high. However I understand what you are after. The NI stuff is a major commitment to get into which is why I have not gone near it. Using a soundcard for the FFT would be obvious. Excel? or Sigview spectrum analyzer - FFT based signal analysis software Sigview may be able to do the whole task. It would be worth an e-mail. If so I'll get a copy as well. And then scrounge a GPIB equipped RF generator.
T&M junky here........
Demian, I have several RF/uWave analyzers with tracking generators here [to 6GHz]. If you are still interested in this approach when i return from Asia at end of summer, we can arrange to get one to you.
The HP is ultra/uber with apc connectors (i have type n adapters for it) See HP 8753D. or a more portable but limited to 1GHz - Instek GSP810. A couple of Anritzu's are lower freq (HF) but dual-channel or diff input.
Would these be of any use?
[PS.. I have a wide band true random (white) noise source.... made by General Radio.]
THx-RNMarsh
Demian, I have several RF/uWave analyzers with tracking generators here [to 6GHz]. If you are still interested in this approach when i return from Asia at end of summer, we can arrange to get one to you.
The HP is ultra/uber with apc connectors (i have type n adapters for it) See HP 8753D. or a more portable but limited to 1GHz - Instek GSP810. A couple of Anritzu's are lower freq (HF) but dual-channel or diff input.
Would these be of any use?
[PS.. I have a wide band true random (white) noise source.... made by General Radio.]
THx-RNMarsh
Last edited:
Hello John,
I hope that the UPD is working again.... maybe this is a tantal cap and not a ceramic?
At work we have a box with tantals (from about 20 years ago) looking right like the UPD ones.
I have done some test automatization with Labview and with Labwindows (the low level C library for sending GPIB commands) in the past.
My personal view is that you are very fast with both approches _if_ you use
it at a daily basis.
For Labview you need the high-level driver for the device. If there are no drivers (=virtual devices) it is a lot more of work.
Another problem is that Labview is a very propriatary approach and
it is not suitable for larger projects. And it costs.
I personally prefer the gpib C approach. Should not be too difficult to write the data to a matlab file for postprocessing.
Basically you need the sequence of GPIB commands for your measurement.
Greetings,
Udo
I hope that the UPD is working again
.... maybe this is a tantal cap and not a ceramic? At work we have a box with tantals (from about 20 years ago) looking right like the UPD ones.
I have done some test automatization with Labview and with Labwindows (the low level C library for sending GPIB commands) in the past.
My personal view is that you are very fast with both approches _if_ you use
it at a daily basis.
For Labview you need the high-level driver for the device. If there are no drivers (=virtual devices) it is a lot more of work.
Another problem is that Labview is a very propriatary approach and
it is not suitable for larger projects. And it costs.
I personally prefer the gpib C approach. Should not be too difficult to write the data to a matlab file for postprocessing.
Basically you need the sequence of GPIB commands for your measurement.
Greetings,
Udo
Udo,
If you ever get a chance to visit me here in Czech then we can run through some Opamp RF IMD tests. I have to perform the test manually, based upon a concept developed by Paul miller.
I did Email you awhile back to see if you would be interested in for a fee writing a test procedure for me using the GPIB interface BUS to automate the tests:-
https://dl.dropboxusercontent.com/u/86116171/Resonances and Repercussions.pdf
Here is more on the subject, ignore my mention of the Labview environment as apparently not even LabView's own Mother loves it:-
http://www.diyaudio.com/forums/equipment-tools/274655-anyone-here-have-experance-ni-labview.html
I meant to mention earlier that the fact the the AES17 filter improves a 20KHz BW limited measurement just highlights the RF non linearity of the test equipment.
If the test equipment had perfect behavior then the filter should have zero impact on the measurement results (presuming the same measurement BW).
It sound not be considered that the filter it artificially "Improving the measurement figures" rather its give more accurate measurement results by reducing the non linearity of the test equipment, and not somehow improving the linearity of the UUT.
I'll post A THD+N plot later - yesterday my UPD started puffing smoke from its side vents, a Ceramic Cap popped making a real mess of the PCB - what a nice Birthday present
https://dl.dropboxusercontent.com/u/86116171/IMG_7352.JPG
Yes, its a decently reliable effect
sometimes it hard to "scope" the RF oscillation.I got caught myself out on my first efforts on a tube design where I'd not used "Grid stopper resistors". There was 10mV ~500MHz oscillation which is very hard to observe on a scope screen at such a low amplitude and frequency. I noticed the noise floor would jump up and down - "the touch test" would trigger the oscillation. Being new to tubes, I was surprised that they would oscillate so high in Frequency - lesson learnt! but I see so many designs without grid stoppers, you just have to wonder....!
For DIY, I was thinking of using a cheap DDS synth IC's from Analogue devices, IIRC they are good to 1GHz
Paul only swept to 200MHz, but with today's world of WiFi / BT & mobile phones 3GHz is a good target.
If you follow EBAY you can get some real bargains - I purchased a HP8664A from Korea shipped to me in Czech Rep for about US$1000 including all import taxes.
Yes indeed, but I wasn't sure how to describe the noise modulation index...I would worry about the PC's cards "tolerance" to RF - some form of Passive LP is advisable which is relevant to this thread - hey maybe even the AUX0025 might be OK
Certainly using "the system" to confirm the PC cards own RF IMD performance would be the first starting point
I hope I don't sound too negative, its not as simple as it seems once to looking into the exact requirements.
My comments about the PC FFT sound card and LPF hold true even for a dedicated Audio FFT analyser.
I've seen chains of inductors used on The HP3561 (Dynamic / FFT analyser) front end PCB - I've not looked closely at the schematic, but wonder if they are part of a passive LPF rather then active LPF which would be a good sign
Paul Miller used the HP3561 in his setup - but he might have added an external LPF.
I don't really know the HP3561, I have one as part of my HP3048A Phase Noise measurement system, but the system is automated via PC/GPIB control so I don't physically need to operate it.
Last edited:
If we could agree on a RF Sig Gen such as a cheap HP8664A
then I'd be happy to support the development costs Agilent/HP 8664A/opt.004, 010 - Performance Signal Generator | eBay
I bought mine unit from this guy, an amazing price considering it has the low phase noise option (004), I guess the 010 option is the weird extra vents on the top lid!
Shipping was surprisingly cheap from Korea (its a heavy unit), but I also got hit by 25% tax here in Czech Rep for a used item.....!
Cheapest 8664A State side on Ebay is this unit:-
http://www.ebay.co.uk/itm/Agilent-H...474?pt=LH_DefaultDomain_0&hash=item3cfae048da
But it does not mention the 004 option (low phase noise) - which is not required for the RF IMD tests, but nice to have for other apps.
Even though HP claim "Low phase noise" with option 004, the HP8664A not in the same league as the HP8662/3
Maybe even consider the HP8662A as I have a couple in the lab, or a HP8642A... both are only good to around 1.2GHz, but they both have far better close in phase noise to the HP8664A.
Go on, go for it
Last edited:
I'll be in HK / SZ from late next week until Aug 13th, anychance your local?
I have a GR unit as well I purchased for these tests and other apps - but I use with a 20KHz AES17 type filter to limit the BW.
Last edited:
Udo,
How could I twist your arm to write the software for me?
good old cash or an AP1? Your relatively "local" so its not to hard to meet
Maybe you and your Girlfriend can visit us in Czech for the weekend and we arrange a local hotel then Renata can take your Girlfriend around visiting the sights while we work on this project together and also test your AP interface etc
Last edited:
Bumping this up. Got myself a Prism Dscope IIIA+ I'm using when repairing/playing around with amps (car audio stuff). As expected, having some problems with class D amps. Seeing higher than expected thdn and some crap aliasing down. Confirmed by playing with some caps at the analyzer input.
Test rig consists of large non inductive power resistors on aluminum bar and a diy server psu supply with current and voltage monitering, good to 60A at 13.5v.
Anyway, not looking to be spoon fed but curious if anyone had any new diy passive implementations rather than splurging on a AP aux-0025 or Prism Sound ds-lpf. Both are completely passive, haven't been able to find any guts pics of either.
Thx
Test rig consists of large non inductive power resistors on aluminum bar and a diy server psu supply with current and voltage monitering, good to 60A at 13.5v.
Anyway, not looking to be spoon fed but curious if anyone had any new diy passive implementations rather than splurging on a AP aux-0025 or Prism Sound ds-lpf. Both are completely passive, haven't been able to find any guts pics of either.
Thx
Last edited:
This is a very interesting topic to me because I'm into measurement.
The best single paper on why a passive low pass filter is required for Class D amplifier measurement was written by the great Bruce Hofer of AP:
Measuring Switch-mode Power Amplifiers
You'll have to register (free) on the AP website to download this paper.
Finally, the one question that is never addressed anywhere that I'm aware of concerns how much attenuation at what frequency is actually required to avoid erroneous readings from the analyzer. AP selected approx. -52dB at 250kHz for the original AUX-0025 and more recently chose the same attenuation at 400kHz for the AUX-0040.
The higher the frequency and/or the lower the attenuation, the less the low pass filter will bias the measurement. Interesting stuff.
The best single paper on why a passive low pass filter is required for Class D amplifier measurement was written by the great Bruce Hofer of AP:
Measuring Switch-mode Power Amplifiers
You'll have to register (free) on the AP website to download this paper.
Finally, the one question that is never addressed anywhere that I'm aware of concerns how much attenuation at what frequency is actually required to avoid erroneous readings from the analyzer. AP selected approx. -52dB at 250kHz for the original AUX-0025 and more recently chose the same attenuation at 400kHz for the AUX-0040.
The higher the frequency and/or the lower the attenuation, the less the low pass filter will bias the measurement. Interesting stuff.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.