Topics

Microwave Noise


Andy G4JNT
 

Until recent years we can safely say all received noise on the uWave bands was purely Gaussian, unlike at HF where its can be moe spiky in nature.

With some of our QRN now coming from Cell Phones and other transmissions, is it still the case the noise can be considered Gaussian?  

What about VHF / UHF ?


Chris Bartram G4DGU
 

An interesting question, Andy, and quite complex to answer. What do we mean by 'noise'?

Gaussian noise is the effect of random (stochastic) processes. The 'noise' generated by radio systems tends to be cyclic, albeit sometimes in a rather complex manner, but is still non-stochastic, particularly after filtering, and is thus not strictly Gaussian!

Professional transmitters - and that includes mobile phone systems - are designed to very high standards with regard to noise outputs, and usually have extensive filtering to allow co-siting of transmitters and receivers. A lot of the 'noise' from other services detected in amateur systems seems to me to be a function of reciprocal mixing, and even-order intermod taking place in our receivers. IMD2 ... is rarely considered in amateur receiver design.

73

Chris G4DGU



Andy G4JNT wrote:

Until recent years we can safely say all received noise on the uWave bands was purely Gaussian, unlike at HF where its can be moe spiky in nature.

With some of our QRN now coming from Cell Phones and other transmissions, is it still the case the noise can be considered Gaussian?  

What about VHF / UHF ?


Andy G4JNT
 

And to add to that, multicarrier systems like OFDM obey all (I think; certainly 'very nearly all') the tests for a Gaussian noise waveform.    Even a comb of a large number pure sinusoids does.

The question really came about after reading an article in the latest QEX about noise monitoring and WSPR, where they use two methods of measuring noise, both of which I've adopted in the past.   The RMS of the time domain signal, measured during the quiet period of no WSPR transmissions.  And the FFT techniques where the bins are put into order of power and noise deemed to be the level in the bin at the 30% point ( I use the lower quartile plus 5dB - thoughts on this differ but the delta is minimal)

In that article they mention that often the two measurement methods give significantly different results, and at other times are very close.   This is put-down to the non-Gaussian nature of noise at HF.    Hence my thoughts at higher frequencies.



On Wed, 16 Sep 2020 at 12:20, Chris Bartram G4DGU <chris@...> wrote:

An interesting question, Andy, and quite complex to answer. What do we mean by 'noise'?

Gaussian noise is the effect of random (stochastic) processes. The 'noise' generated by radio systems tends to be cyclic, albeit sometimes in a rather complex manner, but is still non-stochastic, particularly after filtering, and is thus not strictly Gaussian!

Professional transmitters - and that includes mobile phone systems - are designed to very high standards with regard to noise outputs, and usually have extensive filtering to allow co-siting of transmitters and receivers. A lot of the 'noise' from other services detected in amateur systems seems to me to be a function of reciprocal mixing, and even-order intermod taking place in our receivers. IMD2 ... is rarely considered in amateur receiver design.

73

Chris G4DGU



Andy G4JNT wrote:
Until recent years we can safely say all received noise on the uWave bands was purely Gaussian, unlike at HF where its can be moe spiky in nature.

With some of our QRN now coming from Cell Phones and other transmissions, is it still the case the noise can be considered Gaussian?  

What about VHF / UHF ?


Chris Bartram G4DGU
 

Hello again, Andy,

I can certainly see that off-air HF noise could include artefacts of the non-linear nature of the propagation medium - I'm old enough to remember the 'Luxembourg Effect' on MW! IMD2 does nowadays seem to be taken more-or-less seriously,  particularly since the advent of direct sampling receivers.  But much VHF and up equipment still seems to ignore it.

The use of FFT techniques could easily lead to measurement inaccuracies/uncertainties in a situation where the 'noise' spectrum was non-stochastic. I may be a bit of a dinosaur, but I still use a zero-bias diode detector with proven square-law response, a thermal wattmeter, or alternatively a wideband true RMS voltmeter for making noise power measurements. True RMS voltmeters like the Racal-Dana 9300 or 9303 are available surplus at quite low prices, and with a simple downconverter can be used as a back-end for noise ratio measurements. However that's getting a bit off-topic.

Chris G4DGU


The question really came about after reading an article in the latest QEX about noise monitoring and WSPR, where they use two methods of measuring noise, both of which I've adopted in the past.   The RMS of the time domain signal, measured during the quiet period of no WSPR transmissions.  And the FFT techniques where the bins are put into order of power and noise deemed to be the level in the bin at the 30% point ( I use the lower quartile plus 5dB - thoughts on this differ but the delta is minimal)

In that article they mention that often the two measurement methods give significantly different results, and at other times are very close.   This is put-down to the non-Gaussian nature of noise at HF.    Hence my thoughts at higher frequencies.


Andy G4JNT
 

We're talking  X-purposes.  I'm interested in bandwidths captured by A/D converters, either in a soundcard for narrowish bands or a DS-SDR for wider .   That way it's trivial to obtain an RMS value in the time domain; just square and sum every sample in a defined block length, then root the answer.  Result, exact RMS of that set of samples.   The QEX article does it in blocks of 50ms, then chooses the lowest sum to ensure the highest probability of getting a block with no spurious signals

For the FFT technique, do an FFT on a block, length suited to the frequency resolution of interest and convert the complex number in each bin to a power with Pythagorus.  Then do the ordering and statistics on the result.   That way any signals present get shoved up to the top of the ordered set and can be thrown away.
Practical measuring kit:
A thermal wattmeter will give you true RMS (well, power really) but averaging and time  depends on the thermal mass of the detector and any electronic averaging put in later.   I have a skepticism about "true RMS" meters.  They are pretty universally designed for 50Hz mains, so I suspect their accuracy at audio frequencies until proven otherwise - they probably are OK, but they use an approximation / feedback method to measure RMS, relying on averaging in a capacitor to set a block / time constant for the RMS.   Anyone recall the old Analog Devices AD590?
That's always the trouble with RMS measurement - when is a time period part of the RMS and when is it a varying quantity?   And if a Cap is used in the averaging, where and how is the transition.   (And why is a DC component never included in RMS measurement in any real kit)

Diode detectors may well give you a nice relative value for Sun and Moon peaking purposes, but they drift with temperature, DC offset etc.  Also, if they're calibrated using a carrier then used with noise, there is at least  3:1 ratio for 99% of the noise peak to RMS.    The detector has to remain properly square law to well over 10dB more dynamic range than the thing you're measuring.    OK for Moon noise, but when it comes to sun noise with a large dish and good LNA, you end up needing over 25dB of perfect SQL linearity to make a really accurate measurement.  I remember that problem playing with the FRARS EME system.  The diode detector was a bit suspicious on sun noise measurements.   I believe they've moved to SDR-IQ continuum measurement now

As you may gather - I've become quite a geek when it comes to measuring noise and S/N accurately - it's not an easy subject.



On Wed, 16 Sep 2020 at 14:06, Chris Bartram G4DGU <chris@...> wrote:

Hello again, Andy,

I can certainly see that off-air HF noise could include artefacts of the non-linear nature of the propagation medium - I'm old enough to remember the 'Luxembourg Effect' on MW! IMD2 does nowadays seem to be taken more-or-less seriously,  particularly since the advent of direct sampling receivers.  But much VHF and up equipment still seems to ignore it.

The use of FFT techniques could easily lead to measurement inaccuracies/uncertainties in a situation where the 'noise' spectrum was non-stochastic. I may be a bit of a dinosaur, but I still use a zero-bias diode detector with proven square-law response, a thermal wattmeter, or alternatively a wideband true RMS voltmeter for making noise power measurements. True RMS voltmeters like the Racal-Dana 9300 or 9303 are available surplus at quite low prices, and with a simple downconverter can be used as a back-end for noise ratio measurements. However that's getting a bit off-topic.

Chris G4DGU


The question really came about after reading an article in the latest QEX about noise monitoring and WSPR, where they use two methods of measuring noise, both of which I've adopted in the past.   The RMS of the time domain signal, measured during the quiet period of no WSPR transmissions.  And the FFT techniques where the bins are put into order of power and noise deemed to be the level in the bin at the 30% point ( I use the lower quartile plus 5dB - thoughts on this differ but the delta is minimal)

In that article they mention that often the two measurement methods give significantly different results, and at other times are very close.   This is put-down to the non-Gaussian nature of noise at HF.    Hence my thoughts at higher frequencies.


John Fell
 

G4RFR EME on 10GHz uses a 10GHz HP Spec Ann , on the 618MHz i.f. output of an Octagon LNB ; SDR14 running Spectraview , on a 28MHz down-converted  , Continuum Mode and a G4JNT 144MHz diode detector noise meter system ( 2MHz filter bandwidth)  .All downconversion stages use  GPS referenced LO s .Typical Moon surface noise 1.8dB , Sun at 15.5dB .Echoes with 200W at feed "readable SSB" throughout the Lunar cycle .

Hope to be QRV again soon .

73
John
G0API


On Wed, 16 Sep 2020 at 15:55, Andy G4JNT <andy.g4jnt@...> wrote:
We're talking  X-purposes.  I'm interested in bandwidths captured by A/D converters, either in a soundcard for narrowish bands or a DS-SDR for wider .   That way it's trivial to obtain an RMS value in the time domain; just square and sum every sample in a defined block length, then root the answer.  Result, exact RMS of that set of samples.   The QEX article does it in blocks of 50ms, then chooses the lowest sum to ensure the highest probability of getting a block with no spurious signals

For the FFT technique, do an FFT on a block, length suited to the frequency resolution of interest and convert the complex number in each bin to a power with Pythagorus.  Then do the ordering and statistics on the result.   That way any signals present get shoved up to the top of the ordered set and can be thrown away.
Practical measuring kit:
A thermal wattmeter will give you true RMS (well, power really) but averaging and time  depends on the thermal mass of the detector and any electronic averaging put in later.   I have a skepticism about "true RMS" meters.  They are pretty universally designed for 50Hz mains, so I suspect their accuracy at audio frequencies until proven otherwise - they probably are OK, but they use an approximation / feedback method to measure RMS, relying on averaging in a capacitor to set a block / time constant for the RMS.   Anyone recall the old Analog Devices AD590?
That's always the trouble with RMS measurement - when is a time period part of the RMS and when is it a varying quantity?   And if a Cap is used in the averaging, where and how is the transition.   (And why is a DC component never included in RMS measurement in any real kit)

Diode detectors may well give you a nice relative value for Sun and Moon peaking purposes, but they drift with temperature, DC offset etc.  Also, if they're calibrated using a carrier then used with noise, there is at least  3:1 ratio for 99% of the noise peak to RMS.    The detector has to remain properly square law to well over 10dB more dynamic range than the thing you're measuring.    OK for Moon noise, but when it comes to sun noise with a large dish and good LNA, you end up needing over 25dB of perfect SQL linearity to make a really accurate measurement.  I remember that problem playing with the FRARS EME system.  The diode detector was a bit suspicious on sun noise measurements.   I believe they've moved to SDR-IQ continuum measurement now

As you may gather - I've become quite a geek when it comes to measuring noise and S/N accurately - it's not an easy subject.



On Wed, 16 Sep 2020 at 14:06, Chris Bartram G4DGU <chris@...> wrote:

Hello again, Andy,

I can certainly see that off-air HF noise could include artefacts of the non-linear nature of the propagation medium - I'm old enough to remember the 'Luxembourg Effect' on MW! IMD2 does nowadays seem to be taken more-or-less seriously,  particularly since the advent of direct sampling receivers.  But much VHF and up equipment still seems to ignore it.

The use of FFT techniques could easily lead to measurement inaccuracies/uncertainties in a situation where the 'noise' spectrum was non-stochastic. I may be a bit of a dinosaur, but I still use a zero-bias diode detector with proven square-law response, a thermal wattmeter, or alternatively a wideband true RMS voltmeter for making noise power measurements. True RMS voltmeters like the Racal-Dana 9300 or 9303 are available surplus at quite low prices, and with a simple downconverter can be used as a back-end for noise ratio measurements. However that's getting a bit off-topic.

Chris G4DGU


The question really came about after reading an article in the latest QEX about noise monitoring and WSPR, where they use two methods of measuring noise, both of which I've adopted in the past.   The RMS of the time domain signal, measured during the quiet period of no WSPR transmissions.  And the FFT techniques where the bins are put into order of power and noise deemed to be the level in the bin at the 30% point ( I use the lower quartile plus 5dB - thoughts on this differ but the delta is minimal)

In that article they mention that often the two measurement methods give significantly different results, and at other times are very close.   This is put-down to the non-Gaussian nature of noise at HF.    Hence my thoughts at higher frequencies.


Chris Bartram G4DGU
 

Hello again, Andy,

Yes we are talking cross purposes!

Firstly, I don't disagree with your comments about doing things with an FFT. The maths is maths, and for some kinds of measurement it is absolutely the right approach. But it is, as usual in engineering, a matter of horses for courses.

However, I do disagree with your comments about RMS voltage measurement instruments. The units I quoted are not 'true RMS' multimeters' in the sense that it is used by the manufacturers of general purpose multimeters. I've been caught by their limited bandwidths in the past. The Racal Dana 9300 is a 20MHz BW RMS RF voltmeter, while the 9303 has a bandwidth of 2GHz. I have one of each on my shelves.

I did comment that I was talking about noise power measurements, but perhaps I should have included the word relative in that.

I agree that simple uncompensated diode detectors can, nay will, drift when making level measurements, but their simplicity and wide bandwidth have obvious attractions, particularly for amateur microwave use. Temperature compensation isn't exactly rocket science, either. Extending the measurement range is easily achieved using calibrated attenuators. My approach for noise level difference measurements is to use a calibrated 1dB step
attenuator to set the larger noise voltage incident on the diode to within about 3dB of the smaller, and to read the delta from the detector. And yes, I have measured Sun noise on my old 10GHz dish in excess of 20dB using that technique, albeit on a day when the Sun was rather active!

Do I remember the AD590? I certainly remember using an AD device in a couple of work projects in the late '80s/early '90s which involved the measurement of transmission losses and noise levels at VLF of 2-wire balanced transmission lines lying on the ground with a spacing of 4ft 8 1/2inches ... That may have used a thermal converter IIRC. I also looked at some chips at that time which used analogue multiplication, but they weren't really suitable.

I think I'd agree with you wholeheartedly that noise measurement is something of an art, and getting good accuracy requires attention to detail.

Chris G4DGU


On 16/09/2020 15:54, Andy G4JNT wrote:
We're talking  X-purposes.  I'm interested in bandwidths captured by A/D converters, either in a soundcard for narrowish bands or a DS-SDR for wider .   That way it's trivial to obtain an RMS value in the time domain; just square and sum every sample in a defined block length, then root the answer.  Result, exact RMS of that set of samples.   The QEX article does it in blocks of 50ms, then chooses the lowest sum to ensure the highest probability of getting a block with no spurious signals

For the FFT technique, do an FFT on a block, length suited to the frequency resolution of interest and convert the complex number in each bin to a power with Pythagorus.  Then do the ordering and statistics on the result.   That way any signals present get shoved up to the top of the ordered set and can be thrown away.
Practical measuring kit:
A thermal wattmeter will give you true RMS (well, power really) but averaging and time  depends on the thermal mass of the detector and any electronic averaging put in later.   I have a skepticism about "true RMS" meters.  They are pretty universally designed for 50Hz mains, so I suspect their accuracy at audio frequencies until proven otherwise - they probably are OK, but they use an approximation / feedback method to measure RMS, relying on averaging in a capacitor to set a block / time constant for the RMS.   Anyone recall the old Analog Devices AD590?
That's always the trouble with RMS measurement - when is a time period part of the RMS and when is it a varying quantity?   And if a Cap is used in the averaging, where and how is the transition.   (And why is a DC component never included in RMS measurement in any real kit)

Diode detectors may well give you a nice relative value for Sun and Moon peaking purposes, but they drift with temperature, DC offset etc.  Also, if they're calibrated using a carrier then used with noise, there is at least  3:1 ratio for 99% of the noise peak to RMS.    The detector has to remain properly square law to well over 10dB more dynamic range than the thing you're measuring.    OK for Moon noise, but when it comes to sun noise with a large dish and good LNA, you end up needing over 25dB of perfect SQL linearity to make a really accurate measurement.  I remember that problem playing with the FRARS EME system.  The diode detector was a bit suspicious on sun noise measurements.   I believe they've moved to SDR-IQ continuum measurement now

As you may gather - I've become quite a geek when it comes to measuring noise and S/N accurately - it's not an easy subject.



On Wed, 16 Sep 2020 at 14:06, Chris Bartram G4DGU <chris@...> wrote:

Hello again, Andy,

I can certainly see that off-air HF noise could include artefacts of the non-linear nature of the propagation medium - I'm old enough to remember the 'Luxembourg Effect' on MW! IMD2 does nowadays seem to be taken more-or-less seriously,  particularly since the advent of direct sampling receivers.  But much VHF and up equipment still seems to ignore it.

The use of FFT techniques could easily lead to measurement inaccuracies/uncertainties in a situation where the 'noise' spectrum was non-stochastic. I may be a bit of a dinosaur, but I still use a zero-bias diode detector with proven square-law response, a thermal wattmeter, or alternatively a wideband true RMS voltmeter for making noise power measurements. True RMS voltmeters like the Racal-Dana 9300 or 9303 are available surplus at quite low prices, and with a simple downconverter can be used as a back-end for noise ratio measurements. However that's getting a bit off-topic.

Chris G4DGU


The question really came about after reading an article in the latest QEX about noise monitoring and WSPR, where they use two methods of measuring noise, both of which I've adopted in the past.   The RMS of the time domain signal, measured during the quiet period of no WSPR transmissions.  And the FFT techniques where the bins are put into order of power and noise deemed to be the level in the bin at the 30% point ( I use the lower quartile plus 5dB - thoughts on this differ but the delta is minimal)

In that article they mention that often the two measurement methods give significantly different results, and at other times are very close.   This is put-down to the non-Gaussian nature of noise at HF.    Hence my thoughts at higher frequencies.