Does professional spectrum analyzer such as keysight,RS has `aliasing` problem?

[youta556]

When play SDR equipment,I always change 'sample rate' to avoid `aliasing` problem. 
When I using professional spectrum analyzer,such as keysight n9030a,Should I change 'sample rate' to avoid `aliasing` problem?

[bob91343]

Aliasing is more often an oscilloscope problem.  A spectrum analyzer calls it the way it sees it.

[Marsupilami]

Probably not, but I'm sure it depends on the actual instrument.
In case of such signal analyzers as the 9030 in fft mode those implement an adequate analog IF filter for the maximum sampling rate (or maybe for a couple of lower rates too). That gets rid of the aliasing products during A to D conversion. The bandwidth shown on the screen is further downsampled digitally, again with appropriate digital filtering. This is not particularly complicated but processing intensive especially for high sampling rates, thus lower cost equipment might exclude it and put the burden on the operator to deal with aliasing.
Again I'm not sure but if I had to guess for the n9030a the relevant spec in the datasheet is:

Residues, images, and spurious responses: –100 dBm nominal

Traditional swept spectrum analyzers don't suffer from this issue at all.

HTH

[tggzzz]

Aliasing is more often an oscilloscope problem.  A spectrum analyzer calls it the way it sees it.

You should play with spectrum analysers that don't have a preselector. They usually have a button that helps you visually distinguish signals from aliases.

Tek an HP scopes tended to have optional preselectors, to reduce the cost where they aren't necessary.

[TheUnnamedNewbie]

With the default input, no, I don't think it's common. But with external mixers, it can happen. But then I don't know if you would call it aliasing, it's more a combination of upper vs lower sidebands, and the fact that harmonic mixers mix with all harmonics, and not just the one you really want to measure.

They have some clever tricks to deal with it (like wiggeling the LO a bit, and if you, say, use a 10th harmonic, you expect tones to wiggle 10x that amount. Suppress anything that wiggles a different amount). They all give it fancy names, like Signal ID and so on

[tggzzz]

With the default input, no, I don't think it's common. But with external mixers, it can happen. But then I don't know if you would call it aliasing, it's more a combination of upper vs lower sidebands, and the fact that harmonic mixers mix with all harmonics, and not just the one you really want to measure.

They have some clever tricks to deal with it (like wiggeling the LO a bit, and if you, say, use a 10th harmonic, you expect tones to wiggle 10x that amount. Suppress anything that wiggles a different amount). They all give it fancy names, like Signal ID and so on

The HP8562B doesn't have a pre-selector. The Tek492s have an optional pre-selector. I don't know how many of those were sold, so I can't comment on whether they are common.

You describe the LO wiggling correctly

[Marsupilami]

But with external mixers, it can happen.

The HP8562B doesn't have a pre-selector.

I have a strong urge to pee in the middle of this to assert dominance or something like that, idk. What are you guys talking about?

I was led to believe by the state and the church that aliasing is purely a sampling related problem. When your sample and hold circuit in your A/D converter has more analog bandwidth than 0.5x the sampling frequency. It has nothing to do with mixing, image suppression, oob/harmonic suppression, etc. I feel it was a bit off from OP to call it spectrum analyzer, although it can get somewhat confusing nowadays with combination instruments (e.g. the referred Keysight n9030a), I believe they mean a signal analyzer.

Aliasing is more often an oscilloscope problem.

Bob was kinda close. The inferred instrument (signal analyzer) is effectively a scope after a mixer(s). Aliasing comes from the sampling (scope) part of it, similarly to a cheap zeroIF SDR.

No? I'm not familiar enough with analog instruments so I'm looking forward to learn something new.

[TheUnnamedNewbie]

 When your sample and hold circuit in your A/D converter has more analog bandwidth than 0.5x the sampling frequency. It has nothing to do with mixing, image suppression, oob/harmonic suppression, etc. I feel it was a bit off from OP to call it spectrum analyzer, although it can get somewhat confusing nowadays with combination instruments (e.g. the referred Keysight n9030a), I believe they mean a signal analyzer.

When you think about it, what is a SaH doing when it is sampling a 1 GHz tone with a 700 MHz clock, but mixing it down to a 300 MHz tone? If you were to just ditch everything after that SaH that makes an ADC and ADC, you would end up with a downconversion mixer, usually some kind of passive type with MOS switches, or diode-based.

Aliasing and mixing are not two separate things, they are the same thing, looked at from a different perspective.

[Marsupilami]

When you think about it, what is a SaH doing when it is sampling a 1 GHz tone with a 700 MHz clock, but mixing it down to a 300 MHz tone?

Well, sort of... The difference being is that instead of multiplying (mixing) with a single sine wave we multiply with an impulse train. This is represented by a convolution with an impulse train as well in the frequency domain which makes the spectrum periodic. Aliasing happens when this periodicity is small thus spectral content of different periods overlap. It's just a different phenomenon that doesn't happen in non-sampled systems because the impulse train multiplication never occurs.

[tggzzz]

But with external mixers, it can happen.

The HP8562B doesn't have a pre-selector.

I have a strong urge to pee in the middle of this to assert dominance or something like that, idk. What are you guys talking about?

I was led to believe by the state and the church that aliasing is purely a sampling related problem. When your sample and hold circuit in your A/D converter has more analog bandwidth than 0.5x the sampling frequency. It has nothing to do with mixing, image suppression, oob/harmonic suppression, etc. I feel it was a bit off from OP to call it spectrum analyzer, although it can get somewhat confusing nowadays with combination instruments (e.g. the referred Keysight n9030a), I believe they mean a signal analyzer.

Aliasing is when an out-of-band frequency is folded in-band.

There are several mechanisms by which that can occur. If you intend to restrict your comments to one mechanism, then you should state that.

One of the more interesting mechanisms is where sampling is used to mix frequencies, as in the Tayloe mixer. That's the technology in all the SDR dongles, and elsewhere.

Aliasing is more often an oscilloscope problem.

Bob was kinda close. The inferred instrument (signal analyzer) is effectively a scope after a mixer(s). Aliasing comes from the sampling (scope) part of it, similarly to a cheap zeroIF SDR.

No? I'm not familiar enough with analog instruments so I'm looking forward to learn something new.

Mixing and aliasing is RF-101; there are many tutorials and textbooks available. There's no point us poorly duplicating those.

[Marsupilami]

Aliasing is when an out-of-band frequency is folded in-band.

I might contest that definition. Do you know a widely accepted reference? I'm willing to learn and I'm having difficulties finding THE definition. I'm the product of the digital world but maybe the term was more prevalent in the general sense before the wide spread of sampled systems. Or I just simply missed some of the basics.

The Tayloe mixer is interesting because mathematically the same thing happens there as in sampling. Due to the rectangular nature of the switched "LO" (as opposed to being a sine) the resulting output signal's spectrum will be the convolution of the original spectrum and an infinite periodic spectrum of the switched "LO". The overlaps resulting from this periodicity is what aliasing means in my current understanding.

Mixing and aliasing is RF-101; there are many tutorials and textbooks available. There's no point us poorly duplicating those.

Poorly no. That's why we should yearn to come to consensus about the most accurate and relevant definition, even if it's only referenced. That's how we learn.

[TheUnnamedNewbie]

While

Due to the rectangular nature of the switched "LO" (as opposed to being a sine) the resulting output signal's spectrum will be the convolution of the original spectrum and an infinite periodic spectrum of the switched "LO". The overlaps resulting from this periodicity is what aliasing means in my current understanding.

While we might teach that mixing happens with sines, in practice, for a number of reasons, such as loss and linearity, you often want to drive mixers with a square wave LO, even if it is a single-transistor or diode mixer. This way your 'mixing' non-linear device is in two extreme states, instead of a non-linear region in between.

I don't think you will find a ultimate, end-all definition of aliasing. It is often context dependent. I, looking at things from a millimeter-wave background, think of aliasing as a time-domain view on mixing.

[tggzzz]

I don't think you will find a ultimate, end-all definition of aliasing. It is often context dependent.

Yup

[Marsupilami]

you often want to drive mixers with a square wave LO

Ok, that's a valid point. If I use the Tayloe mixer as an example then the concept is true for any system where the LO has significant harmonic content. I still feel like that those are rather called harmonic products or harmonic spurs rather then aliasing anything.

[JohnG]

FWIW, as an undergrad some decades ago, the term aliasing was used in an analog sense to describe how an undesired image would appear in the same band as a desired signal. I would guess that the first usage of the term was pre-digital, though I have been surprised before.

Also, sampling is just another form of mixing. It is just a question of what the harmonic distribution of the LO is. Sine wave LO has the fundamental only, square wave has all odd harmonics falling off as 1/n, impulse train has all harmonics of equal amplitude.

John

[Marsupilami]

FWIW, as an undergrad some decades ago, the term aliasing was used in an analog sense to describe how an undesired image would appear in the same band as a desired signal. I would guess that the first usage of the term was pre-digital, though I have been surprised before.

I understand what you guys are saying I just can't find any material about it. I was schooled in the early 2000s, so I'm a spring chicken compared to a lot of microwave veterans here. By that time basically everything was digital and I can imagine the use of the term got somewhat biased toward that kind of interpretation, however I'm through pages and pages of google search results as well as checking a few textbooks (relatively recent ones though, e.g. Pozar's Microwave Engineering) and whitepapers from manufacturers like Marki etc. and I found absolutely zero mention of aliasing in a context unrelated to sampling. (Either 1D for ADCs and such or 2D for image processing) But I have to admit that I haven't found anything that would conclusively rule out the more general interpretation either.

Also, sampling is just another form of mixing.

I know that I'm pushing it, but I think this also depends on how do you define mixing (and/or sampling for that matter). If you define it as the time domain multiplication of two (or more) signals then maybe, but mostly (correct me if I'm wrong) mixing refers to multiplication resulting in frequency translation, which is not happening in sampling unless the signal is (purposefully or not) undersampled.
You mentioned the frequency characteristic of the square wave (or imperfect LO) and the ideal impulse train. The difference might be regarding nomenclature is which component in the mixing signal's spectrum is purposeful and which are unwanted side effects. In case of (baseband) sampling really what does the meaningful job during multiplication is the DC component of the sampling clock (leaves the original signal where it was), anything else is causing unwanted spectral content. For mixing (as with the purpose of frequency translation) the purposeful component can be anything but DC.
If one tried to (re)define aliasing from sampling in analog language it could be something like: Inseparably spectral overlap due to spectral periodicity in input of time domain multiplication. And from that it could be easily concluded that there's nothing specific in it to sampling. I'm not against even wider definition of it that applies to any kind of spectral overlap I just can't find it in literature, but now I'm more curious then ever to discover some.

[TheUnnamedNewbie]

I know that I'm pushing it, but I think this also depends on how do you define mixing (and/or sampling for that matter). If you define it as the time domain multiplication of two (or more) signals then maybe, but mostly (correct me if I'm wrong) mixing refers to multiplication resulting in frequency translation, which is not happening in sampling unless the signal is (purposefully or not) undersampled.

You don't get to choose whether or not you look at it from timedomain or frequency domain. They are the same thing. If you have two signals going into a non-linear device (which a S/H is) one of the things that will come out is the product.

This does, in fact, happen in a S/H. IF you were to just look at the spectrum after the S/H without a anti-aliasing filter, it will contain these frequency components too - its just that usually the anti-aliasing filter and he input spectrum is such that they don't occur (or do by design in the case of RF ADCs/DACs).

I just want to be clear that I wouldn't use aliasing for spurs or anything. I understand that aliasing is usually used for a specific concept with ADCs, and similarly, I wouldn't call a S/H a mixer or vice versa. I just want to point out that, conceptually, mathematically, they are the same. A S/H and mixer can be thought of almost like an I-beam and an H-beam. Different table in the catalogue, different rules, but conceptually the same thing, just with the 'trade-off sliders' put at a different place.

[Marsupilami]

You don't get to choose whether or not you look at it from timedomain or frequency domain.

I'm not. I called out multiplication in the time domain to make it clear which operation I'm thinking of because it is not equal to multiplication in the frequency domain.

IF you were to just look at the spectrum after the S/H without a anti-aliasing filter, it will contain these frequency components too

Yes, but. First you wouldn't put an antialiasing filter after your S/H during AD conversion because that invalidates the hold part of it, but I understand your point, and yes, I agree you would see the repeating harmonic components.
Second the concept of sampling does not rely on the use of an actual S/H circuit. You could imagine a magic device that, without any interference, especially multiplication, would create the discrete time domain representation of the analog signal. Assuming this purely mathematical relationship the periodicity in the spectrum of the sampled signal would still be present.
What I'm trying to point out is that it appears to me that in contemporary nomenclature aliasing refers to the non-recoverable spectral overlap of sampled systems that is their property because they are discreet in time and not because of their physical implementation.
I would agree with you about the mathematical equivalence if it didn't have this property. You can easily imagine an ideal mixer that doesn't have the spectral periodicity but you can't imagine a sampled signal without it. I'm not holding against you if you think this is tomato-tomato but in my opinion it is a enough of a difference to be the basis of classification whether or not the term aliasing is used to describe the phenomenon.

[ejeffrey]

Since aliasing in an ADC is just mixing products with clock and it's harmonics aliasing and mixing are at least closely related.

In the context of a benchtop spectrum analyzer you can't distinguish them anyway unless you know the exact receiver topology used for the current frequency and span settings so there is little point making a distinction.
 

[Marsupilami]

Since aliasing in an ADC is just mixing products with clock and it's harmonics

I don't think that is true. I understand why it is a useful representation. It makes dealing with the clock non-idealities such as frequency drift / phase noise / jitter super convenient. You handle it as mixing (regardless whether the actual sampling system physically performs such a function or not) then assume that an ideal circuit is sampling the signal after the mixer. But that's not what is happening.
You start with a continuous time signal that has an aperiodic spectrum, discard a whole bunch of (potential) information (everything between two samples) and you'll end up with a discrete time signal that has a strictly periodic spectrum and this has nothing whatsoever to do with the fact that in practice you needed a sampling clock to facilitate this transformation. This is just how discrete time signals are.

Assuming an ideal pulse train as the sampling clock, when you mix it with the input signal what you do is multiply the signal value by 1 at every sampling moment and by 0 for the rest of the time. Then when your ideal sampler operates on this mixed signal it is going to discard all the information that was already zeroed out by the mixing. But if you didn't do the mixing the ideal sampler would still discard everything in between the sampling moments resulting in the very same discrete time output.

[JohnG]

The ideal sampling operation is mathematically equivalent to multiplying the signal with an ideal periodic impulse train. It doesn't matter how you implement the multiplier, and even if you don't show it, it is still there because that is the sampling process by definition. It is the multiplication that is important.

John

[Marsupilami]

The ideal sampling operation is mathematically equivalent to multiplying the signal with an ideal periodic impulse train. It doesn't matter how you implement the multiplier, and even if you don't show it, it is still there because that is the sampling process by definition. It is the multiplication that is important.

I know that I'm getting into ridiculous levels of philosophical nitpicking and it's also a looks, swims and quacks like a duck problem but this will still keep me up at night.
The equivalence you mention between the discrete time signal and the continuous time signal multiplied by the impulse train is through their spectrum. If you perform a Fourier transform on the impulse train multiplied continuous time signal (continuous time->continuous frequency domain ) and a discrete-time Fourier transform on the discreet time signal (discrete time->continuous frequency domain) those results will be equal. This equality creates the connection between the continuous time and the discrete time signal. However note that the impulse train multiplied continuous time signal and the discrete time signal are still not equal.
Those two functions operate on a different domain, one being f: R->R while the other g: Z->R and the mapping between those two cannot be done by R*R multiplication. The spectra can be equal because we used two different kinds of Fourier transforms on the two different domain functions.
For every possible practical purpose this is totally irrelevant, I'm with you on that. However what I'm still trying to emphasize is the philosophy behind the unwanted spectral changes in a sampled signal. By sampling you're discarding (potential) information. The primordial cause for any resulting artifacts in the output is this loss of information and not that it was multiplied by another signal, even though you can derive the same result from representing the whole process with such multiplication. And because of this, while understanding the similarities I think there is a fundamental difference in the unwanted spectral overlap of a mixer and an ADC even if they quack the exact same way.

[JohnG]

Well, it's been a long time (decades) since I had to do anything involving communication or signal theory, so it's a bit of a stretch to remember the math.  However, where you see a fundamental difference, I see a fundamental similarity. The circuits that do the work are just implementing the math, and if there are two or more solutions, they are mathematically equivalent. The one you pick is the one that is most practical to implement with current technology (or at least the technology one knows how to implement). But, I tend to see similarities across disciplines in any case.

The ideal mixer building block is a multiplier. But, the "LO" input to the mixer does not have to be a sine wave. It could be a square wave, which is commonly used. Once you allow for a non-sinusoidal periodic waveform with infinite harmonic content, it becomes pretty arbitrary to allow square waves but not an impulse train. Once you go to an impulse train, you have created an ideal sampler. The discrete time signal has to come from sampling a signal somewhere, and if it has frequency content >Fs/2, it means the image overlaps the desired content, and that is the origin of aliasing error.

John

[tggzzz]

Well, it's been a long time (decades) since I had to do anything involving communication or signal theory, so it's a bit of a stretch to remember the math.  However, where you see a fundamental difference, I see a fundamental similarity. The circuits that do the work are just implementing the math, and if there are two or more solutions, they are mathematically equivalent. The one you pick is the one that is most practical to implement with current technology (or at least the technology one knows how to implement). But, I tend to see similarities across disciplines in any case.

The ideal mixer building block is a multiplier. But, the "LO" input to the mixer does not have to be a sine wave. It could be a square wave, which is commonly used. Once you allow for a non-sinusoidal periodic waveform with infinite harmonic content, it becomes pretty arbitrary to allow square waves but not an impulse train. Once you go to an impulse train, you have created an ideal sampler. The discrete time signal has to come from sampling a signal somewhere, and if it has frequency content >Fs/2, it means the image overlaps the desired content, and that is the origin of aliasing error.

I'm on "your side" here, however I note that sometimes aliasing is required for system operation.

Consider an old analogue sampling scope, the ones with the hypnotic twinking display that videos never capture. Those take one sample per sweep, which is fine for stationary signals (and most signals can, with appropriated skill and imagination, be made stationary )

Thus I have a Tek1502 which measures 140ps risetimes while sampling at ~40kS/s. Without aliasing of the sampling signal, it wouldn't work.

[Marsupilami]

However, where you see a fundamental difference, I see a fundamental similarity. The circuits that do the work are just implementing the math, and if there are two or more solutions, they are mathematically equivalent. The one you pick is the one that is most practical to implement with current technology (or at least the technology one knows how to implement). But, I tend to see similarities across disciplines in any case.

I don't debate the similarity, I'm just biased towards one of the equivalent descriptions to be more right. I see though, that my stance is not strong enough to be able to claim that I'm right in an objective sense. I find it interesting that how our interpretation of what is important about the explanation of the phenomena differs based on our background. Honestly going into this whole debate I was expecting that finding exact, conclusive definitions would be easier and about that I was clearly wrong.

Thus I have a Tek1502 which measures 140ps risetimes while sampling at ~40kS/s. Without aliasing of the sampling signal, it wouldn't work.

I worked with analyzers having a... I don't really know what's the proper name, sampled superhet? architecture. In these the first stage mixer downcoverts to an IF that is set to be in the middle of the 2nd or 3rd Nyquist band of a following digitizer.
E.g. 50MHz wide band was donwconverted from RF to between 100MHz-150MHz and then sampled with an ADC clocked at 100MHz. In this case the concept what you guys are arguing is even more apparent because the 2nd stage downconversion to baseband is clearly performed by undersampling. In my mind that still is sort of a nice side effect of process but again I understand that might not be the universal truth.