Measuring Noise Figure


richard@...
 

Hi all


I watched a Youtube video recently about measuring noise figure using the measured gain of an amplifier and a spectrum analyser. The thing that I found odd was that the video maker did not terminate the input of the amplifier - he said it did not matter. I can think of two reasons why it might matter:


1) the input load resistor provides the base thermal noise level (I suppose it might not matter much if the amplifier is very noisy)

2) some amplifiers are not entirely stable when un-terminated.


Would you agree?


73 Richard G3CWI


Derek Kozel <derek.kozel@...>
 

There are different approaches to measuring noise figure. I always use a y factor test and leave the noise source attached to the input of the amplifier/receiver. Powering on and off the noise source provides the change in stimulus to calculate the noise figure but keeps the termination as close as possible, including reducing thermal changes in the short term.

On Mon, Jul 18, 2016 at 7:09 AM, richard@... [ukmicrowaves] <ukmicrowaves@...> wrote:
 

Hi all


I watched a Youtube video recently about measuring noise figure using the measured gain of an amplifier and a spectrum analyser. The thing that I found odd was that the video maker did not terminate the input of the amplifier - he said it did not matter. I can think of two reasons why it might matter:


1) the input load resistor provides the base thermal noise level (I suppose it might not matter much if the amplifier is very noisy)

2) some amplifiers are not entirely stable when un-terminated.


Would you agree?


73 Richard G3CWI



Chris Trask <christrask@...>
 

Thoroughly disagree. I use the 3dB method, which requires a 50-ohm termination, a 3-dB attenuation pad, a noise source, and a means for measuring output noise power (typically a spectrum analyzer). You first measure the amplifier output noise with the 50-ohm termination at the input. Then you add the 3-dB attenuator to the output and the noise source to the input. You adjust the noise source power until the two measurements are the same, then go through some calculations to determine the NF.


I watched a Youtube video recently about measuring noise figure using the measured gain of an amplifier
and a spectrum analyser. The thing that I found odd was that the video maker did not terminate the input
of the amplifier - he said it did not matter. I can think of two reasons why it might matter:

1) the input load resistor provides the base thermal noise level (I suppose it might not matter much if
the amplifier is very noisy)
2) some amplifiers are not entirely stable when un-terminated.

Would you agree?


Chris

When the going gets weird, the weird turn pro
- Hunter S. Thompson


richard@...
 

"Thoroughly disagree"  Thanks for your reply Chris. I am not clear with which approach you thoroughly disagree with however.

a) the gain method of measuring noise figure
b) measuring without an input load
c) measuring with an input load; or possibly
d) Derek's approach.

73 Richard G3CWI


Martin - G8JNJ
 

Hi Chris,

Your method sounds interesting - I'd like to try it.

Can you outline the calculations required, or point to a site or source that has the detail ?

Regards,

Martin - G8JNJ

www.g8jnj.net
www.tc2m.info
http://websdr.suws.org.uk


Chris Trask <christrask@...>
 

Yes. When I get home I'll locate a PDF document I wrote for doing it.


Hi Chris,

Your method sounds interesting - I'd like to try it.

Can you outline the calculations required, or point to a site or source that has the detail ?

Regards,

Martin - G8JNJ

Chris Trask
N7ZWY / WDX3HLB
Senior Member IEEE
http://www.home.earthlink.net/~christrask/


Chris Trask <christrask@...>
 

B


"Thoroughly disagree" Thanks for your reply Chris. I am not clear with which approach you thoroughly disagree with however.

a) the gain method of measuring noise figure
b) measuring without an input load
c) measuring with an input load; or possibly
d) Derek's approach.


73 Richard G3CWI

Chris

When the going gets weird, the weird turn pro
- Hunter S. Thompson


Richard GD8EXI
 

Any noise measurement you make with an open transmission line is going to be dependant on its length.

Rather than me try to explain the theory I suggest you try the following experiment.

Connect a good 50 ohm dummy load to your 432MHz or above receiver under test and note the noise level (reading 1). Now disconnect the dummy load and take a series of readings (readings 2) with different lengths of open transmission line attached. Easily done by adding adaptors or short lengths of coax. You should find reading 2 is dependant on the length of the open transmission line. Finally add a dummy load to the end of your adjustable length transmission line to confirm you get the same value as reading 1.
 
If you are looking for a minimum equipment noise estimate I suggest you put your dummy load in the deep freeze for a few hours and while still very cold connect it to your receiver and note the noise level. Now let it warm to room temperature and note the reading again. You are looking for a very small change in noise level and indeed you might need to heat your dummy load to about 100C to see an easy to read difference. This is the hot and cold dummy load method and it should be reasonably accurate as long as the impedance of your dummy load does not change much with temperature.
 
I would be interested in what readings people get.
 
Richard
GD8EXI




On 18/07/2016, 18:19, "richard@... [ukmicrowaves]" <ukmicrowaves@...> wrote:

 
 
 
   

"Thoroughly disagree"  Thanks for your reply Chris. I am not clear with which approach you thoroughly disagree with however.

a) the gain method of measuring noise figure
b) measuring without an input load
c) measuring with an input load; or possibly
d) Derek's approach.

73 Richard G3CWI
 
   



Andy G4JNT
 

I did it once with boiling water and room temperature on what was supposed to be a low noise amp (144MHz) of the late 1980s era.  I saw  a genuine "fraction of a dB" but back then didn't have the means to measure mean noise power reliably or accurately, especially in a wide bandwidth.

Simplistically:
Assume a 1dB noise figure LNA with sufficient gain to overcome any following stages.
1dB is equivalent to a noise temp T of  75K   (  since NF = 10.LOG(T / 290 + 1)

So the delta noise power seen at the output of the system for input at 100C = 373K and  Room Temp 290K 
= :
10.LOG[ (373 + 75) / (290 + 75) ]  = 0.89dB

Which ought to be measurable with modern wide bandwidth SDRs.  It wasn't reliable using a diode on the output of an SSB Rx in 1988 !

Andy  G4JNT

On 19 July 2016 at 11:54, Richard Baker perwick@... [ukmicrowaves] <ukmicrowaves@...> wrote:
 

Any noise measurement you make with an open transmission line is going to be dependant on its length.

Rather than me try to explain the theory I suggest you try the following experiment.

Connect a good 50 ohm dummy load to your 432MHz or above receiver under test and note the noise level (reading 1). Now disconnect the dummy load and take a series of readings (readings 2) with different lengths of open transmission line attached. Easily done by adding adaptors or short lengths of coax. You should find reading 2 is dependant on the length of the open transmission line. Finally add a dummy load to the end of your adjustable length transmission line to confirm you get the same value as reading 1.
 
If you are looking for a minimum equipment noise estimate I suggest you put your dummy load in the deep freeze for a few hours and while still very cold connect it to your receiver and note the noise level. Now let it warm to room temperature and note the reading again. You are looking for a very small change in noise level and indeed you might need to heat your dummy load to about 100C to see an easy to read difference. This is the hot and cold dummy load method and it should be reasonably accurate as long as the impedance of your dummy load does not change much with temperature.
 
I would be interested in what readings people get.
 
Richard
GD8EXI




On 18/07/2016, 18:19, "richard@... [ukmicrowaves]" <ukmicrowaves@...> wrote:

 
 
 
   

"Thoroughly disagree"  Thanks for your reply Chris. I am not clear with which approach you thoroughly disagree with however.

a) the gain method of measuring noise figure
b) measuring without an input load
c) measuring with an input load; or possibly
d) Derek's approach.

73 Richard G3CWI
 
   




Christopher Bartram <cbartram@...>
 

Like Andy, a long while ago (late '70s) I also used the hot/cold load method to estimate NF. In my case at 144, 432, and 1296MHz. As they for the purpose of my measurements they were relatively easy to define physically, I used the temperatures of the triple-point (ie. effectively melting ice) and boiling water. All good standard school physics lab. stuff.

The load - which remained connected - was an 1/8W metal film resistor de-leaded and suitably waterproofed. No change of return-loss was detectable over the 0 - 100C range.

My noise receiver at that time was a linear 2ishMHz-wide IF strip with a square-law detector at 29MHz preceded by a 2m converter. It was easy to check the linearity of the detector, while the assumption that the noise spectrum at the amplifier output didn't change between the two source temperatures seemed reasonable - and still forms part of the assumptions made by modern, high-end NF-ometers.

That system was sufficiently accurate to confirm the measurements of the 144MHz amplifiers which I'd designed made with a commercial saturated thermionic diode noise source based NF-ometer. This unit effectively used the same technique as Chis Trask outlined. The performance of the commercial unit wasn't good enough to make realistic measurements at higher frequencies, but I was able to get a friend, working in the research labs of a semiconductor company, to measure my other amplifiers, and there was also good agreement.

A few months later, I brought (brand-new!) an HP346 source. That still forms the basis of my NF measurement kit, and is still producing readings which correlate well with other sources/instruments to beyond 10GHz - including modern Keysight demonstrators. Nowadays I use the source with an HP8790B. The home-brew IF strip still exists, and is just about to to be used as part of a new microwave EME system in order to enable accurate cold sky/Sun, cold sky/Moon and cold sky/ground measurements.

The hot/cold source technique is still probably the only way of making realistic measurements, and I'm very dubious indeed about noise figure '_measurement _' systems which don't use some form of calibrated source.

If you want to learn more about NF measurement, I very strongly suggest a careful read of the HP application note written by Ian White, GM3SEK and David Stockton, GM4ZNX.



Vy 73

Chris
GW4DGU


Richard GD8EXI
 


I have recently did the experiment I suggested back in July with a hot and cold dummy load, using a Signal Hound SA44B with 0.32dB ±0.16 NF  (noise temp 22K ±11) 40dB gain LNA connected between it and the dummy load. It took a temperature raise from –10 to +65C to see ~1db a raise in the recorded noise level.
The results are shown in the graph below.  The hot and cold dummy load method giving a LNA noise temp of ~ 51K so not very differ considering the length of the extrapolation needed. The conclusion being it works as Andy G4JNT said it should.

I did the experiment because was trying to measure Sky noise on the higher bands and wanted to test the accuracy of my equipment.

73s
Richard
GD8EXI



   

I did it once with boiling water and room temperature on what was supposed to be a low noise amp (144MHz) of the late 1980s era.  I saw  a genuine "fraction of a dB" but back then didn't have the means to measure mean noise power reliably or accurately, especially in a wide bandwidth.

Simplistically:
Assume a 1dB noise figure LNA with sufficient gain to overcome any following stages.
1dB is equivalent to a noise temp T of  75K   (  since NF = 10.LOG(T / 290 + 1)

So the delta noise power seen at the output of the system for input at 100C = 373K and  Room Temp 290K 
= :
10.LOG[ (373 + 75) / (290 + 75) ]  = 0.89dB

Which ought to be measurable with modern wide bandwidth SDRs.  It wasn't reliable using a diode on the output of an SSB Rx in 1988 !

Andy  G4JNT

On 19 July 2016 at 11:54, Richard Baker perwick@... [ukmicrowaves] <ukmicrowaves@...> wrote:

 
 
 
   

Any noise measurement you make with an open transmission line is going to be dependant on its length.

Rather than me try to explain the theory I suggest you try the following experiment.

Connect a good 50 ohm dummy load to your 432MHz or above receiver under test and note the noise level (reading 1). Now disconnect the dummy load and take a series of readings (readings 2) with different lengths of open transmission line attached. Easily done by adding adaptors or short lengths of coax. You should find reading 2 is dependant on the length of the open transmission line. Finally add a dummy load to the end of your adjustable length transmission line to confirm you get the same value as reading 1.
 
If you are looking for a minimum equipment noise estimate I suggest you put your dummy load in the deep freeze for a few hours and while still very cold connect it to your receiver and note the noise level. Now let it warm to room temperature and note the reading again. You are looking for a very small change in noise level and indeed you might need to heat your dummy load to about 100C to see an easy to read difference. This is the hot and cold dummy load method and it should be reasonably accurate as long as the impedance of your dummy load does not change much with temperature.
 
I would be interested in what readings people get.
 
Richard
GD8EXI




On 18/07/2016, 18:19, "richard@... <http://richard@...>  [ukmicrowaves]" <ukmicrowaves@... <http://ukmicrowaves@...> > wrote:

 
 
 
   

"Thoroughly disagree"  Thanks for your reply Chris. I am not clear with which approach you thoroughly disagree with however.

a) the gain method of measuring noise figure
b) measuring without an input load
c) measuring with an input load; or possibly
d) Derek's approach.

73 Richard G3CWI