Some spectrum analyser questions from a student

[finos]

Hello everyone!
EE student here dealing with RF BlackMagic! Last day we visited our RF lab measure the characteristics of a RF amplifier.
The exercise was for us to design and -if safe- implement our design and get data.

I had to measure
1. True gain and compare it to the specs given.
2. Harmonics and “wasted” power 3. True power output. 
4. Linearity.

“Easy” I thought! Ive done my Homework, ive seen my share of eevblog & theSignalPath videos and I understand my way around these instruments.
My design was a Spectrum analyzer with TG (or an external signal gen), attenuators so I don’t blow the heck out of our R&S FSV 40Ghz, and our amp. Id connect the TG on the amp input, the output on a attenuator calculated with ehough headroom to not cause damage but not too large, to lower the signal much so the instument could not "see" it properly .
The professor gave me the green light and fired up the machines, made my measurements and uploaded the report.

I got the feedback a couple of minutes ago

The values you measured and wrote where well within spec for the exercise, but I am skeptical regarding your technique. I would not trust the power output reading from a spectrum analyzer,Typically we use the N1914A, a dedicated RF power meter, for that purpose.

First of all, I barely knew the lab was equipped with Agilent / R&S / Anritsu gear, going into this exercise I thought we are going to “play” around with some rigol or siglent gear.
Second, and here comes the question for you guys:

I understand that a dedicated instrument must have better accuracy but why is the spectrum analyser readings not trust worthy (especialy of its grade)

Can you spot Something wrong my test setup?
I set span, bandwidth and center frequency according to the Amps specs, then I used TxTotal and Peak search to get the dbm for the power output & calculate the gain taking into account the 47db attenuator I chose.

I would ask him that but we wont be meeting or having any lessons with him again for 2 weeks so I came here (Sir , if you are reading (everyone is on the eevblog forum, so there is a chance..) this please don’t take back my passing grade


For completion sake i attach a photo of the setup.

ps

I know what are you thinking , i thought the same.
The coffee and water next to the instrument belonged to our professor. And yes, I panicked at the thought of spillage, too.

[radiolistener]

A spectrum analyzer measures the power of individual frequency components within a signal, while a power meter measures the total power across all frequencies. This is the key difference between the two.

When measuring the power of a specific frequency component with a spectrum analyzer, you exclude the power of other components. However, for an amplifier - especially one operating under overload conditions or in a nonlinear region - harmonics can be significant. These harmonics contribute to the total power, but they are omitted when analyzing only a single frequency component.

[Solder_Junkie]

The “standard” test for linearity is to use two signals of the same amplitude and measure the level of 3rd, 5th, etc intermodulation products.

In the case of a linear amp, the signals would be two carriers through a ferrite combiner. For an SSB transmitter you would use audio tones.

See attached showing the IMD products from a typical 12 Volt amateur SSB transceiver fed with 2 audio tones. The other image shows the significant improvement using pre-distortion techniques with the same transceiver.

SJ

[MarkT]

RF spectrum analyzer accuracy for levels is not great.  Thats even after self-calibration.  RF circuitry and signal paths are full of losses, components with frequency responses that are not flat and both level- and temperature-dependent.  Relative measurements of level at very similar frequencies are good, but absolute measurements much less so.   S.A's have a ton of components in the signal path, compounding the accuracy issue. Dedicated RF power meters are frequency and temperature agnostic and have very short signal paths, and are thus more accurate/stable, some use thermal heating to measure power I think which can be calibrated against a DC source for highest accuracy, but you still have the variation in frequency reponse of attenuators and cables to worry about.

Accurate measurement of absolute level at RF is always hard as everything has (frequency-dependent) loss, cables, attenuators, connectors (that's why you use a torque-wrench on metrology grade RF connectors, and why they have a very short calibrated lifetime).

[JoanBS]

Why are there 5 attenuators in cascade at the input of the amplifier?

[Bud]

Politely tell your professor that a power meter measures the combined power at the frequency of interest, harmonics, intermodulation components, spurs, and noise across the entire frequency range. Whereas a spectrum analyzer can be used to measure the useful signal at the frequency of interest, or any other those components individually. Therefore a SA is more accurate for the task of measuring useful power (providing that specifications on accuracy for both the power meter and SA are comparable).

Also, move 3 or 4 attenuators from amplifier input to output. This will improve output match across the amplifier frequency range, improving accuracy of the characterization process.

[G0HZU]

FWIW here's my view on this.

The amplifier is being driven at a very low level because the output on the analyser appears to be -49dBm. So there is little risk of any distortion terms from the amplifier. The test frequency is 868 MHz so there is going to be some loss and mismatch uncertainty caused by the long output cable.

The best way to measure the output power accurately will usually be with the power meter as long as you keep several dB away away from the 1dB compression point.

That analyser doesn't look to be a high end lab instrument, so it may not have a low input VSWR at 868 MHz. If the input VSWR was 1.2:1 and the source match of the amplifier (via the output cable) was 1.8:1 then the mismatch uncertainty here is about +/- 0.23dB. The loss in the cable might be 0.3dB and the overall uncertainty in that mid range analyser might be in the ballpark of +/- 0.3dB. So this all adds up to a lot of uncertainty.

A better way to measure the output power accurately would be to connect the power meter head direct to the amplifier output (providing the amp can't overload and damage the power head). Then use fewer attenuators at the input and run the amplifier up to a higher output level, maybe 0dBm (assuming the P1dB is >10dBm)

The typical power sensor VSWR might be 1.03:1 at 868 MHz so the mismatch uncertainty here is +/- 0.04dB (rather than +/- 0.23dB with the mid range analyser).
The uncertainty for the power meter + sensor would typically be less than 0.2dB. You could also use the power meter to measure the drive level that gets fed to the input of the amplifier. This should also give a result with low uncertainty.

So if you use the power meter for the gain and the output measurements (as your professor suggested), you should get lower uncertainty for the gain and the absolute power level assuming you aren't driving the amplifier close to compression and generating lots of harmonic distortion within maybe 15dB of the fundamental.

That's hardly likely to be happening with your current setup at -49dBm output...

[G0HZU]

Why are there 5 attenuators in cascade at the input of the amplifier?

It does seem strange, but I think the Hittite sig gen has a limited output level range. It looks like it can't go much lower than -15dBm at the output. Also, it can generate several hundred milliwatts output power so it's wise to have attenuation at the output.

So I can understand why there might be more than one attenuator at the output of the sig gen. Five attenuators does seem like a lot though

[JoanBS]

Well, it seems there are actually only four attenuators, but I'm not entirely sure about this supposed amplifier...

The RF generator is set to -15 dBm; let's assume the four attenuators are 10 dB each, so at the amplifier input we would have (-15 dBm) -40 = -55 dBm, and the SA shows a signal of -49.70 dBm. If we subtract about 0.70 dB of losses in cables and connectors, we get a total of 6 dB of gain in the amplifier!

That seems like very little gain for an amplifier. Also, the SA signal looks suspiciously clean, with no trace of harmonics of any kind, so I'd say the amplifier is operating well below its normal level, which would invalidate the actual power measurement with the SA.

But it would be necessary to know the values ​​of the attenuators for my conjectures to be correct...

[G0HZU]

Well, it seems there are actually only four attenuators, but I'm not entirely sure about this supposed amplifier...

The RF generator is set to -15 dBm; let's assume the four attenuators are 10 dB each, so at the amplifier input we would have (-15 dBm) -40 = -55 dBm, and the SA shows a signal of -49.70 dBm. If we subtract about 0.70 dB of losses in cables and connectors, we get a total of 6 dB of gain in the amplifier!

That seems like very little gain for an amplifier. Also, the SA signal looks suspiciously clean, with no trace of harmonics of any kind, so I'd say the amplifier is operating well below its normal level, which would invalidate the actual power measurement with the SA.

But it would be necessary to know the values ​​of the attenuators for my conjectures to be correct...

The carrier frequency is 868 MHz and the span on the analyser is only 1 MHz so there won't be any harmonics visible on such a tiny span. However, the power level on the analyser is only -49dBm. There's lots of reasons why the professor would critique that measurement method with the analyser (vs a correctly used power meter) and I've tried to list a few of them in my previous post.

[pdenisowski]

The values you measured and wrote where well within spec for the exercise, but I am skeptical regarding your technique. I would not trust the power output reading from a spectrum analyzer,Typically we use the N1914A, a dedicated RF power meter, for that purpose.

I understand that a dedicated instrument must have better accuracy but why is the spectrum analyser readings not trust worthy (especialy of its grade)

If you're going to measure power in a non-frequency-selective way, then a power sensor / meter is the way to go.  Sensors are also less expensive than a comparable spec an, which also makes it less painful if you accidentally fry one

The accuracy of the FSV (assuming it's in cal) shouldn't be an issue.  The biggest issue I've seen in measuring power with a spectrum analyzer is how many people measure it. 

Simply placing a marker on the signal and reading off the value is okay in some applications, but the "best" way to measure power with a spec an is to use zero span mode.  Assuming you have a wide enough resolution bandwidth setting, this should give you an acceptably accurate measurement of RF power in many applications.

I tested this using an FSW (next step up from FSV) and an NRP18S power sensor on a CW signal.  The difference between the power sensor measurement (-29.92 dBm) and the zero/span channel power measurement (-29.72) was 0.2 dB.

And yes, I panicked at the thought of spillage, too.

I like to live dangerously ... see attached

[Edit: note that the picture shows a "marker" measurement, not a zero span or channel power measurement]

[pdenisowski]

Hi @G0HZU  First, thanks for your previous posts! I agree with everything except this

That analyser doesn't look to be a high end lab instrument, so it may not have a low input VSWR at 868 MHz.

The FSV is the next tier down from our "flagship" FSW analyzer and is widely used in many, many labs  

You do make a very valid point about the effect of VSWR matching on power levels, but you would / could have the same issue with a power sensor.  We actually provide an uncertainty calculator for use with our power sensors, and matching plays a big role here too (see sections 3.3.2 - 3.3.4 of the linked paper)

https://cdn.rohde-schwarz.com.cn/pws/dl_downloads/dl_application/application_notes/1gp_psp/Primer_Power_Sensors_V2.pdf

[G0HZU]

Hi @G0HZU  First, thanks for your previous posts! I agree with everything except this

That analyser doesn't look to be a high end lab instrument, so it may not have a low input VSWR at 868 MHz.

The FSV is the next tier down from our "flagship" FSW analyzer and is widely used in many, many labs  

You do make a very valid point about the effect of VSWR matching on power levels, but you would / could have the same issue with a power sensor.  We actually provide an uncertainty calculator for use with our power sensors, and matching plays a big role here too (see sections 3.3.2 - 3.3.4 of the linked paper)

https://cdn.rohde-schwarz.com.cn/pws/dl_downloads/dl_application/application_notes/1gp_psp/Primer_Power_Sensors_V2.pdf

OK but the FSV is still not what I would class as a high end spectrum analyser, it's probably at the upper end of mid range. A quick glance through the datasheet will show this isn't a high end analyser. Even R&S describe it as an analyser for the cost conscious.

It is still a very, very nice analyser though


It's also classed as mid range here back in 2008:
https://www.microwavejournal.com/articles/15616-fsv-signal-analyzer


The VSWR spec is <1.5:1 on the low band (up to 3.6 GHz), with 1.3:1 typical so I went with 1.2:1 in my worked example.

A high end analyser would probably have an input VSWR spec of <1.2:1 on the lowest band and it would typically be much better than that.
Also a high end analyser will have better phase noise, lower spurious, better linearity and better dynamic range.

It may be the case that the input VSWR is actually much better than the datasheet implies, but it does state 1.3:1 typical on the low band.

[G0HZU]

The other thing to note is that the screen appears to show a RBW of 10 kHz and a span of 1 MHz. The displayed noise floor is down at -105dBm so there can't be much in the way of front end attenuation selected.
If the analyser noise figure is typically about 22dB near 1 GHz then the displayed noise floor with a 10 kHz RBW will be -174 + 40dBHz + 22dB  = -112dBm with no RF attenuation at the front end.

So there probably isn't even 10dB attenuation selected. The screen is fuzzy, but it looks like the attenuation is a single digit amount.

The FSV analyser only offers 5dB steps for the front end attenuator although 1dB steps is an option. So I'm going to guess that the front end attenuation setting is only 5-8dB depending on what options are fitted to it.

I'd expect your professor may have spotted all this and hopefully he will agree with my comments

[G0HZU]

The biggest issue I've seen in measuring power with a spectrum analyzer is how many people measure it.

Simply placing a marker on the signal and reading off the value is okay in some applications, but the "best" way to measure power with a spec an is to use zero span mode.  Assuming you have a wide enough resolution bandwidth setting, this should give you an acceptably accurate measurement of RF power in many applications.

I tested this using an FSW (next step up from FSV) and an NRP18S power sensor on a CW signal. 

Can you explain why you think zero span offers an advantage when measuring a single cw (unmodulated) signal as in your case above? I'm aware that it can be used for measuring signals that are modulated, eg pulsed or gated as long as averaging is used with the average power detector enabled. It can work really well with signals that have some form of modulation.

I don't see the realistic advantage (of using zero span mode) with an unmodulated cw signal compared to making a peak search with a swept display and a sensible choice of RBW. It would be different if the test signal was modulated and occupied some bandwidth.

Also, your analyser setup below looks to be quite poor. There's a few potential issues if the aim was to make accuracy comparisons between a power sensor and a spectrum analyser at 868 MHz. I can advise how to set it up better if that helps?

[pdenisowski]

Can you explain why you think zero span offers an advantage when measuring a single cw (unmodulated) signal as in your case above? I'm aware that it can be used for measuring signals that are modulated, eg pulsed or gated as long as averaging is used with the average power detector enabled. It can work really well with signals that have some form of modulation.

I don't see the realistic advantage (of using zero span mode) with an unmodulated cw signal compared to making a peak search with a swept display and a sensible choice of RBW. It would be different if the test signal was modulated and occupied some bandwidth.

Indeed.  I wasn't sure if the OP's signal was CW or not, so I suggested zero span because it's the "safest" way of measuring power with a spec an.  Assuming RBW is "close" to the (non-zero) width of the nominally CW signal, then you should get essentially the same value using a marker as in zero span. (And it doesn't even matter which detector type you use )   

I've had many customers try to measure non-CW power with a marker, so my "standard" recommendation is always to use zero span.  It even works on noise alone

Also, your analyser setup below looks to be quite poor. There's a few potential issues if the aim was to make accuracy comparisons between a power sensor and a spectrum analyser at 868 MHz. I can advise how to set it up better if that helps?

(laughs)  The picture was just a "graphic" and a bit of a joke (since the OP pointed out the drink very close to the analyzer in their picture) - it wasn't meant to be a guide as to how to set up highly accurate power measurements on a spec an

[G0HZU]

OK thanks.

I can list a few issues with the setup of your FSW analyser to clarify what I mean. If the aim is to minimise measurement uncertainty to make critical measurements then it's really easy to 'bake in' extra uncertainty beyond what is shown on the datasheet and I can see a few examples of this below.

It's generally not a good idea to select 0dB attenuation at almost any time using a spectrum analyser because it degrades the input VSWR and the overall measurement uncertainty.

For critical amplitude measurements, using FFT mode on a fairly wide span is usually inferior to swept mode when it comes to the measurement uncertainty for any signals contained in that span. Swept mode has the advantage that all measurements take place at the same (calibrated) IF centre frequency for each IF.

I'm not familiar with the FSW and how it sets up FFT measurements and I'm not sure how relevant the 1001 points message is on the screen but it is usually very risky to use 1001 points on a 1 MHz span with a 2 kHz RBW when in FFT mode. I don't know what window function the FSW is using or how many FFT sub bands it is using but there could be scalloping losses with your setup that will contribute significant uncertainty that will depend on the precise frequency of the test signal. 1 MHz divided by 2 kHz is 500 and you only have 1001 data points. That's OK for general signal analysis but not for making fairly precise amplitude measurements in FFT mode. It just adds yet more uncertainty.

I would recommend using a decent attenuator at the far end of the test cable to improve the source match. I can't quite see past the coke can but I don't think you have an attenuator there.

The analyser isn't fully warmed up according to the onscreen message but maybe this message was about to disappear anyway


I'm fairly certain that I've not used the FSW model although the company I work for did get a flagship 26 GHz analyser from R&S maybe 12-15 years ago to evaluate. I can't remember the model number

[G0HZU]

It might be interesting to show what a classic old 'high end' analyser can do in terms of input VSWR on band 0 where band 0 covers from LF through to 2.5 GHz or maybe 3.5 GHz for some analysers.

First up is my old HP 8566AB spectrum analyser. This is over 40 years old now and is an A version that has had the factory conversion to a B. Maybe the very latest high end analysers from the top manufacturers can compete in this VSWR test but I suspect they still can't...  This is because this analyser was designed almost without regard for cost or size or power or weight. The HP 8566 was the industry standard spectrum analyser for over two decades and some aspects of its performance are impressive even today.

The s11 plot below is very old but it shows a worst case return loss of about 28dB across band 0 (with 10dB attenuation selected) where band 0 stops at 2.5 GHz. The plot extends past this to 3 GHz.

This is a VSWR of better than 1.08:1 across LF through 2.5 GHz with 10dB attenuation selected. If the attenuation is increased to 20dB the VSWR improves to a worst case of 1.05:1 across the frequency range. This is a sample of one and there will be some spread from analyser to analyser, but this performance level is hard to beat...

[G0HZU]

The successor to the HP8566B was the Agilent E4440A PSA analyser and this was introduced some time around 2001 as Agilents flagship spectrum anayser. I've got one here and it is 20 years old now and I've owned it for about 6 years now. It hasn't been formally calibrated at Keyight since about 2018.

See below for a VSWR plot of my PSA analyser on band 0 across 0-3 GHz with 10dB and 18dB attenuation selected.
The PSA is at its best with about 18dB attenuation selected at the input but it is still OK at just 10dB. This PSA has seen a lot of hard use and I can't guarantee that the input attenuator hasn't been stressed by a previous user. It may have been better than this when it was new.

The other plot below shows the typical amplitude response of the PSA on band 0. This is supplied by Keysight and overlays the typical input frequency response across six examples of the PSA analyser. This is remarkably good, but it requires a test source with an ultra low source VSWR to reproduce these results.

The PSA is quite dated now in terms of the display and the digital IF and DSP that it contains but the front end design of the frequency converter is still very impressive. You can see that it isn't quite as good as my old HP 8566AB but maybe I should measure the 8566 again as the plot for the 8566AB is probably 20 years old now and I could use a better VNA to measure it.

This is the level of performance I would expect from a 'high end' lab analyser.

[G0HZU]

The natural successor to the PSA is the N9040A/B UXA analyser.  There is also the 9030A/B PXA analyser but the front end design of the PXA isn't quite in the same class.

The plot below is for the current N9040B UXA analyser. Sadly I don't own one of these but this is the analyser I would upgrade to (maybe one day  ) if the PSA failed BER and I couldn't get another PSA analyser.

The input VSWR is fairly similar to the PSA on band 0. I would really like to own a UXA analyser one day....

[G0HZU]

If you explore beyond band 0 and start looking at test frequencies of >10 GHz, then the difference between using a power meter and a spectrum analyser becomes much more significant, even if a $$ high end analyser is used. A decent power sensor should still have an input VSWR of < 1.05:1 at frequencies up to about 12 GHz. A spectrum analyser will struggle to maintain an input VSWR below 1.4:1 above 10 GHz.

If the DUT source match was 1.8:1 and the analyser SWR was 1.4:1 then the mismatch uncertainty is about +/- 0.41dB when using a typical lab analyser at 10 GHz.
If the DUT source match was 1.8:1 and the power sensor SWR was 1.05:1 then the mismatch uncertainty is about +/- 0.06dB when using a decent power sensor at 10 GHz. This is a huge improvement compared to using a spectrum analyser.