EMRFD Message Archive 1790

Message Date From Subject
1790 2008-07-09 17:08:15 Jason Milldrum Noise Figure Measurement without a Spectrum Analyzer

I've been looking into what it takes to make noise figure measurements
on amplifiers, mixers, etc. here in the shack, and it seems that
pretty much everyone assumes that you have a spectrum analyzer at your
disposal. Since I don't yet have a SA handy, is it feasible to do this
measurement using a power meter instead of a SA? I'm guessing that one
of the big advantages of using a spectrum analyzer is the dynamic
range and the amplification. However, with the W7ZOI power meter (and
similar instruments like the M3 Electronix FPM-1), you have a pretty
good dynamic range for cheap. The amplification shouldn't be hard to
do either.

So a block diagram like this be something that I could use to make
decent measurements?
Noise generator ---> DUT ---> Amplifiers (MMIC?) ---> Filtering ---> Power Meter

Seems like it would work, but I feel like I might be missing
something. Is this workable, and if so what other considerations do I
need to make?

Thanks for your help,
Jason Milldrum, NT7S
Amateur Radio Station NT7S - <http://www.nt7s.com>
NT7S Blog - <http://www.nt7s.com/blog>
qrp-l.org Flickr Group - <http://flickr.com/groups/qrp-l/>
1791 2008-07-09 18:18:36 kerrypwr Re: Noise Figure Measurement without a Spectrum Analyzer
An SA is not mandatory.

Have a look at some of the HP/Agilent appnotes;


AN-57 is very good; it describes the measurement methods.

Others such as AN-57-1 are more detailed.

Maxim also has a good appnote;


Sabin W0IYH's Oct 92 QST article is another good read.

I am building a version of the Measurement Receiver; one of its uses
will be to make good NF measurements. A noise figure meter as made by
HP and others is really a receiver with known parameters, eg IF noise
1792 2008-07-09 18:49:30 Jason Milldrum Re: Noise Figure Measurement without a Spectrum Analyzer

1793 2008-07-09 18:52:14 Jason Milldrum Re: Noise Figure Measurement without a Spectrum Analyzer
Arggh, please disregard that last e-mail. My notebook PC is going haywire.

Thanks very much for pointing me in the direction of those application
notes. The AN-57 series looks like exactly what I need!

Jason NT7S

1795 2008-07-10 13:39:02 Wes Hayward Re: Noise Figure Measurement without a Spectrum Analyzer
Hi again Jason,

Yes your block diagram should work just fine. You need enough gain
ahead of the power meter to be able to see the noise in the BW of
interest. You probably want to use a bandpass filter that covers a
ham band. If it is wide, it is easier to get noise measurements.
But you want it to be narrow to get a "spot noise figure"
measurement. Be sure that the filter performs well in attenuating
frequencies well away from the band of interest. The way to do this
is to put the bandpass in cascade with a low pass filter, ideally
built with small components that still work well at UHF.

I often measure the NF of receivers. I merely run them without AGC
and use an audio volt meter on the output. If you then put an
amplifier ahead of the receiver, you can measure the composite NF and
find the NF of the amp alone.

If you do the measurements in a narrow bandwidth receiver, such as
the ones we use for SSB or CW, you will see a lot of fluctuation in
the audio voltmeter. You may have to add some video filtering to
guarantee that you are seeing an average in the meter.

You are
1796 2008-07-11 02:59:00 Jason Milldrum Re: Noise Figure Measurement without a Spectrum Analyzer
Hi Wes,

Thanks for verifying that this method will work. Those Agilent
application notes that were mentioned earlier were very helpful in
understanding the theory and process behind the measurements.
Regarding the filter, I understand that if your filter is too wide
(wider than the BW of the DUT), then you run the risk of getting an
inaccurate measurement due to the non-linear frequency response of the
DUT. However, if the filter BW is too narrow, you need to take an
average measurement in order to maintain accuracy. I imagine that it's
not going to be hard to to keep my filter BW narrower than the BW of
the BF998. However, do you have any suggestions on a lower BW limit
that I should avoid in order to get good accuracy?

Your method of using the DUT as a preamp for a receiver sounds very
interesting and might be a bit easier to do. Could you possibly
elaborate on it a bit more, please? Mainly, I'm curious about how you
would calculate the composite NF and the DUT NF based on the audio
voltmeter readings.

That brings me to one more question, if you don't mind. From what I
have read, in order to measure noise figure, I need to know the ENR of
the noise source. I have an Elecraft noise gen and homebrew zener
noise gen. How do I go about characterizing the ENR of these sources.
This is one area which isn't totally clear to me from reading the
Agilent documentation.

Jason NT7S

1797 2008-07-11 03:56:13 kerrypwr Re: Noise Figure Measurement without a Spectrum Analyzer
I've just remembered that the Sabin paper on sensitivity measurement
is #52 on the EMRFD CD; it covers most/all of your questions.

Sabin's article on building a noise source (#53 on the CD) is also
very useful; it addresses both building & actual use of the source.

Noise/Com no longer calibrate amateur-built noise sources; I was very
fortunate to have mine calibrated elsewhere.

But you don't need a calibrated source; look at the second method (the
"gain method") in the Maxim appnote and read the Sabin paper.
1798 2008-07-11 07:35:18 Gary Johnson Re: Noise Figure Measurement without a Spectrum Analyzer
Some good news: Noise Com is calibrating sources again (as least for US customers). This
is from an email forwarded to me by Bill Sabin:

As you are well aware, in business terms 1994 was a long time ago. Noise/Com
to this day continues to offer Amateur Radio Experimenters discounts on
diodes. Calibrations of these Noise Sources have been more a factor of
available resources. After review of your request, on a trial basis we will
resume our offer to calibrate these noise sources at a substantial discount.
Pricing will need be $100.00, turn around time will be a function of
available resources and will most likely be quoted at 8 - 12 weeks. This
will be for domestic customers only, as we have had too many issues with
noise sources being held up in customs. Customers will need to contact
Noise/Com for an RMA (Return Merchandise Authorization).

Best Regards,
Al Sebolao
Sales Manager
25 Eastmans Road
Parsippany, NJ 07054
(973) 386-9696 Voice
(973) 386-9191 Fax

-Gary, WB9JPS
1800 2008-07-11 13:59:16 Wes Hayward Re: Noise Figure Measurement without a Spectrum Analyzer
Hi Jason and group,

Yes, your scheme should be fine. But as you point out, the filter
that is at the tail end of the measurement chain should have a
bandwidth that is more narrow than the amplifier you are testing.
The fluctuation that is removed by averaging is hardly observable in
a bandwidth of 10 kHz or more. When measuring the NF of a SSB
receiver, or other noise, I see fractions of a dB in a SSB bandwidth,
so that is not a big problem. Things do get worse with a CW
filter. You then want that video filtering.

You know, now that I think of it, I am not sure about the statistics
of measurement so far as the AD8307 power meter goes. I have
certainly seen situations with it where I was seeing noise. However,
I don't know how accurate it is. That is a place for some serious
comparisons. The AD8307 does some unexpected things when subjected
to a two tone signal, as was pointed out by Bob Kopski (K3NHI).

Regarding the noise figure of a cascade: This is standard stuff.
You will find the formulas in section 2.6 of EMRFD. Also, it is
easy to derive from the fundamental definitions. There is
information on the intercepts of cascades in EMRFD section 6.3.

If you get started and do a measurement, you can then compare your
calculations with those from the program "Cascade.exe" that is on the

When using a receiver for the NF measurements, you start by measuring
the NF of the receiver. You then add the DUT and do a measurement
of the cascade. You must also measure the gain of the DUT.
Having this data, you can then calculate the NF of the amplifier
alone. This works fine even if you have loss instead of gain. For
example, you should be able to confirm that the NF of a 6 dB pad is
indeed 6 dB. You have to remember when doing these calculations
that the formulae are all in terms of raw power ratios rather than dB
values, so conversions are required.

Calibration is always an issue. I calibrated my noise source by
borrowing a noise source that had been used in another lab. Or I
might have taken my noise source to that lab. I would have to look
back at my notes. The calibrated noise sources that we started with
were calibrated by the manufacturer of the noise figure measurement
system, which was HP. Then the sources were periodically re-

Years and years ago, I participated in some after hours experiments
with wa7tzy where we built our own hot-cold source. This consisted
of using two different 50 Ohm resistances. One was at room
temperature of about 295 K while the other resistor was immersed in
liquid nitrogen at 77K. The available noise power from each was
then well defined. The complication came when we looked back at
the resistors with a network analyzer. Most resistors changed R
quite a bit when cooled to LN temperatures. But we found some that
were fairly stable. This work happened in the mid 1970s when we
were building an EME system for 432 MHz. As I recall we ended up
with about a 1.5 to 1.7 dB NF for our receiver. That is lousy by
today's standards, but it was pretty good for that time.

My own noise source is built for an ENR of about 22 dB. (ENR is
excess noise ratio.) This is more than you would want for most
applications, but I wanted the ability to measure the NF of switching
mode mixers with loss. An ENR of around 6 to 8 dB is a better
general purpose tool. You get this merely by picking the
attenuation of a pad that follows the noise source. That pad is
then kept in the noise source including the time when it is

There are some other schemes for noise measurement that can be found
in older issues of QST. In one, a suitable light bulb is turned on
to a controlled level. The bulb is chosen so that it will have an
internal resistance of 50 Ohms when it is on. The color temperature
is measured with optical methods that can perhaps be implemented with
some physics labs. Some study will be required to dig out the
details. Anyway, if you know the color, you know the temperature,
and you then know the available noise power.

In the past when I didn't have a calibrated source available, I used
a couple of schemes to "fake it." These are not recommended
measurement methods, but they are better than nothing at all. In
one case, I had noticed that a dual gate MOSFET that was set up for
as a high gain amplifier would come in with a NF around 1 dB. The
gate 1 would be driven with about a 3K impedance at the measurement
frequency, realized with an L-network. So I just built such an
amplifier, measured it with a homebrew noise source of unknown ENR.
I would then assume NF of 1 dB and calculate ENR to match.

In another case I would carefully measure the noise bandwidth of a
receiver. This requires good frequency readout resolution. Once
this is known, you measure the sensitivity of the receiver with a
good signal generator. In that era, I used a homebrew crystal
controlled source with a step attenuator. The homebrew crystal
oscillator had been calibrated against the best signal generator I
could find around Tektronix at the time, which was (you guessed it)
an HP-8640B. My source was in SSD and is in Fig 7.31 of EMRFD. The
original appearance was July 1975 QST.

Do not look at a receiver "spec" that says you are using a 2100 Hz
crystal filter and assume that to be the receiver bandwidth. Audio
peaking can alter things a lot. You really do have to measure the
noise bandwidth. The details of this calculation are
1801 2008-07-11 16:17:18 Glen Leinweber Re: Noise Figure Measurement without a Spectrum Analyzer
I'd heartily endorse going through the Agilent application notes on
measuring noise: AN-57-1 is absolutely superb. Others in this
series are too.
Support for older appnotes like these seems to be dropping away -
Agilent used to mail free hard copies.
Archives of some classic old appnotes (like AN 57-1) are available:

With steep-sided filters, effective noise bandwidth is often estimated
as the -3dB points for flat-topped filters. The EMRFD measurement
receiver (Fig. 7.60) has a gaussian-topped filter, about 250 Hz. wide.
Any idea how the gaussian top affects noise bandwidth?
With such a small bandwidth crystal filter, the log amp output may
need a longer time constant low-pass filter to smooth out variations
in noise measurements.
From the AD8307 data sheet, I see that noise measurements vs.
sine measurements may be in error by +0.5 dB. I think all these log
detectors (spectrum analysers too) make similar errors. So calibrating
noise system gain with a RF signal source requires correction when
switching to noise.

As bandpass filter bandwidth gets narrower, I'd guess that noise
crest-factor reduces toward that of a sine wave. For wideband noise
you might want to leave 10 dB overhead for noise peaks in your
measurement setup. For sine waves you only need 3dB overhead.
Has anyone seen a relationship between bandwidth and crest factor,
or am I barking up a wrong tree?
You DO have to ensure linearity in ANY noise measurement setup.
For example, using a DVM to measure AC noise coming out of a
receiver's audio may give non-linear readings because noise peaks
are sliced off (especially if you're reading near the top of a range).
1802 2008-07-11 17:31:30 Wes Hayward Re: Noise Figure Measurement without a Spectrum Analyzer
Hi Glen and gang,

Yes, the old HP applications notes are great. Bravo to Agilent for
putting them on the web.

A narrow filter like that Gaussian to 6 design that you built will
indeed require some smoothing after the log amp. Fortunately, that
is a really easy thing to do; 1 RC and an op-amp and you are there.

What you need to do with a non square top filter is to plot the thing
and calculate an integral. Get your data as a voltage that is a
function of frequency. The effective noise bandwidth is then
(1/(Vp^2))*integral(V(f)^2)*df where the integral is over all
frequency from f=0 to large value. In practice, all that is
required is to go from points that are 20 or 30 down on either side
of the peak. Vp is the peak voltage and V(f) is the voltage as a
function of frequency. See Equation 6.1-11 from IRFD, p 209.

Actually, if you had a receiver with a CW crystal filter that had a
reasonable clean response, the noise bandwidth would be pretty close
to the 6 dB down value. But many receivers have just enough peaking
in the audio that you do not get good correspondence between MDS and
NF measurements. On the other hand, if you do the math and measure
the actual response, you can come amazingly close.

The error that you have with a spectrum analyzer log amp, and
essentially any log amp, is just the result of the mathematical
crunching that happens. In my prior statement I was just thinking
(well, typing) out loud and wondering about the details of the 8307.
I should have dug out the data sheet.

It's amazing how much mileage we, as experimenters, have gotten from
that chip. Kudos to Barrie for that effort.

My gut feeling is that the relati
1803 2008-07-11 18:36:21 kerrypwr Re: Noise Figure Measurement without a Spectrum Analyzer
The noise bandwidth of a Gaussian filter is, according to Willtek;


approx. 1.2 times the resolution (3 dB) bandwidth.

The Gaussian-to-6 (or 12) is not a "full" Gaussian but the 1.2 figure
gives a "sanity check".

I'm not very "mathematical" but I'm very "graphical" so I used a
graphical method to estimate the NBW of the Gaussian-to-6 filter I
built for my Measurement Receiver.

When I finally achieved a lovely smooth response I converted the dB
response figures to microvolts. I then used Excel to plot the
response on linear/linear scales and printed it.

I then overlaid a grid of 10 mm squares on the print.

This is an age-old and surprisingly accurate way to measure an
irregular area.

Count-up whole squares and estimate part-squares within the curve;
10%, 50% etc.

Add all these areas to find the area under the curve; divide this by
the height of the peak.

The result is the width of a rectangle with height equal to the curve
peak; this is the NBW.

In my case I had a filter with -3 dB width of 215 Hz; the noise
bandwidth by the above method is 230 Hz.

Not too far from the 1.2 figure and smaller as expected (steeper
skirts) so my method passes the "sanity check".
1806 2008-07-12 01:54:10 Jason Milldrum Re: Noise Figure Measurement without a Spectrum Analyzer
Excellent, I've printed these out and will peruse them over the weekend.

Jason NT7S

1808 2008-07-12 15:04:17 rotfunkblau Re: Noise Figure Measurement without a Spectrum Analyzer
Hello together,

for those interested in simple receiver data measurements Elecraft
offers some mini modules (noise gen, signal generator, ...).
If this might be an opti
1809 2008-07-12 15:04:22 rotfunkblau Re: Noise Figure Measurement without a Spectrum Analyzer
Hello together,

for those interested in simple receiver data measurements Elecraft
offers some mini modules (noise gen, signal generator, ...).
If this might be an opti
1810 2008-07-12 15:04:26 rotfunkblau Re: Noise Figure Measurement without a Spectrum Analyzer
Hello together,

for those interested in simple receiver data measurements Elecraft
offers some mini modules (noise gen, signal generator, ...).
If this might be an opti