At 07:29 AM 10/7/2007, Rob Sherwood wrote: >What a treat to have someone with this kind of background >contributing on the subject. >Here is how I measure phase noise on any radio. I realize it is not >perfect, but most radios have enough phase noise, it seems to work. >When compared to a 1960s or 70s radio (Drake C-Line or Collins >S-Line) that have band crystals an a PTO, modern radios are much >worse at least between 1 and 80 kHz.
That's the thing about PLL sources.. You get that flat plateau from the reference noise floor. And, if the "rules" only require spurious signals on transmit to be down X dB, there's not much incentive to try and make it better. And if the transmit spectra noise is only down X dB, there's not much incentive to make the receiver a whole lot better, because you'll be limited by the transmit sidebands on everyone else's radio. So this tends to push towards everyone having about the same performance, and that performance is wherever a basic competent design with non-exceptional components will get you. Especially in the ham market, where users tend to keep their radios for decades, the "need" for better performance will be limited, because the band is still full of moderately noisy transmitters from the 70s and 80s. Interestingly, when you get regulatory or operational requirements that push harder, the designs improve (look at the NTIA redbook requirements for government radios. Most run of the mill ham rigs can't meet the spectral mask, even if they meet the frequency accuracy.. on the other hand, ALL of the commercial radios meet it, because they have to.) >The generator I use are either an HP 8662A or an HP 8642A. All >measurements are on 20 meters only, as with everything else I >publish. I measure the noise floor using a 2.4 kHz filter. If I >can select a 1 kHz bandwidth I may use that, and some times I run >the numbers at 500 Hz as a double check. I then place the generator >a 10, 20, 40 and 80 kHz, and increase the level until the HP 3400A >RMS meter increases 3 dB on the speaker output, just like one does >to measure noise floor. Sometimes the radio will show blocking >tendencies at 10 kHz, making that data point invalid. Let me sort of repeat this back to make sure I understand.. You have the radio tuned to, say, 14.1 kHz, and you measure the noise floor at the audio output. Then, you put the signal generator at 14.120 kHz, and run the level up until the signal increases by 3dB. There's a combination here of the filter attenuation in the radio (i.e. the IF skirt) and the reciprocal mixing with LO phase noise. I haven't looked much at ham receiver performance in detail, but what's a reasonable assumption for IF ultimate attenuation close in? Say you have a 6 pole filter with a bandwidth of 2kHz. At 10 kHz away, you'd expect it to be 120 dB down, but can you really get that much ultimate attenuation. Maybe the flat "adjacent signal rejection" is due to the filter characteristic, rather than reciprocal mixing? A digital receiver might well have a very unusual stop band characteristic.. certainly it won't have the usual flat slope down forever to the noise floor like you see with an analog filter. If one is using the radio in SSB mode, you could distinguish by using a very frequency selective voltmeter (or digitizing and FFTing, which is what you do with a phase noise test set), because the LO phase noise contribution is at the apparent tuned frequency (14.1) as opposed to the IF leakage (which is 20 kHz away). If one changed the buffer length in PowerSDR, I think that would change the ultimate attenuation floor of the "IF" filter, assuming that the filter is implemented using all the samples in the buffer (i.e. if it's a frequency domain FFT, or a time domain equivalent). That is, it's the FFT noise floor you're seeing, not the LO phase noise. So, any F5K owners out there that want to do a quick check.. tune 20 kHz away from your clean crystal source, and see what the level of the source changes as you change the buffer length. >Generally the data shows a drop of about 6 dB per octave as one >moves out from 10 to 20 to 40 to 80 kHz. The Orion is an exception, >as its LO noise peaks around 50 to 60 kHz out, by design. If I >cannot get a good reading at 10 kHz, and the other three data points >are consistent, then I extrapolate the published 10 kHz number. Of >course I correct for bandwidth, 30 dB for a 1 kHz filter, 27 dB if I >used a 500 Hz filter, or 4 dB for a 2.4 kHz filter. I consider this >measure my least accurate, likely at +/- 3 dB. Because of the difference between the integrated noise power bandwidth vs the stated 3dB bandwidth? > Gerald said he expected much higher numbers on the 5000A, the way > they divide down the LO, thus the comment on the possible cause. I > have no idea the cause. In general, the phase noise for a NCO is going to be a combination of: 1) The reference noise - 20 log10( fref/fnco) 2) phase accumulator roundoff (can be made negligible by carrying lots of bits) 3) phase truncation into the lookup table roundoff 4) dac precision (analog quantization) 5) broadband noise from the DAC buffers 6) aliases and IM from things like glitch energy that wind up inband 7) nonlinearities in the dac #2,3, and 4 can be reduced by techniques like error feedback (which I think the new Analog Devices parts use), which also helps suppress close in spurs. #5,6,7 are also pretty low in modern devices So I'd actualy expect the noise of the NCO output to be quite good, assuming the reference oscillator is quiet. Gerald may be excessively pessimistic. The overall analysis of all these contributors is fairly complex, and in the SDR1000/F5K world, you also have to throw in the effects of the QSD and audio digitizer. Lorenzo Simone at Thales Alenia Space-Italia has a nice paper at the 2003 Aerospace conference that talks about the NCO noise stuff and breaks out a lot of the factors involved in a synthesized RF chain. I'll see if I can find a way to put a copy up for public consumption (it's copyright IEEE). If you have access to IEEE Xplore, you can google for "Simone Alenia X/X/Ka Transponder" and it will turn up a bunch of links to the paper. If anyone has a reference to a detailed analysis for the QSD, I'd love to see it (so I don't have to grind through it myself, being fundamentally lazy....<grin>) _______________________________________________ FlexRadio mailing list [email protected] http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz Archive Link: http://www.mail-archive.com/flexradio%40flex-radio.biz/ FlexRadio Knowledge Base: http://kb.flex-radio.com/ FlexRadio Homepage: http://www.flex-radio.com/
