RF preamp circuit selection

[G0HZU_JMR]

It does seem to make a big difference. I plotted the same results using the Tek analyser when set to similar settings. This analyser has a much lower noise floor yet it shows the same profile for both cases. Both analysers have a much lower noise floor than either case and this comparison helps to show that the analyser noise isn't affecting either set of results.

Unfortunately the Tek RTSA doesn't display noise as nicely as the Agilent PSA on these settings and I had to use a lot of averaging with the Tek analyser.

The Agilent PSA produced the earlier plot using a classic swept first LO and a digital RBW filter.
The Tek does the whole thing with a fixed LO and an FFT and lots of averaging.

This partly explains why the Tek trace doesn't have the same 'noisy' look to it. However, the average noise levels are the same so it helps to show the results are valid.

[G0HZU_JMR]

It's probably also worth mentioning that the input impedance of the SA602A is not the same as what is stated in a typical datasheet. Most datasheets show a diagram of a 1500R shunt resistor at pins 1 and 2 and the datasheet states the input impedance is 1500R in parallel with about 3pF.

However if you measure the DC voltage at pins 1 and 2 it is typically shows 1.4V. If there really were 1500R resistors to ground as shown in the datasheet diagrams then this would mean about 1mA in each pin and this alone would use up nearly all of the stated 2.4mA current consumption for the entire chip. So that diagram in the the datasheet is clearly wrong. I suppose it could be argued that the diagram shows the AC equivalent of grounding the input with a 1500R resistor but that is wrong too.

This is because the SA602A typically has a (single ended) input impedance of about 3500R in parallel with 2.5pF when measured on a VNA across HF and into VHF. Note that the VNA power has to be turned down really low to prevent a false reading through circuit overload. I measured it a while back and got the plot below and I think this used a VNA source power of -40dBm.

This shows that the input impedance is more like 3000R in parallel with 2.5pF in the HF to low VHF region. Whilst it might seem to be a bold claim to state that the SA602A datasheets have been incorrect for about 35 years it is possible to do some digging and find a Philips app note AN1993

The direct link to AN1993 is below:

https://www.nxp.com/docs/en/application-note/AN1993.pdf

In section 4.1 it states that the input impedance of the SA602A is about 3k || 3pF and this does not agree with the SA602A datasheet or the SA602A internal circuit diagram presented above it. However it DOES agree with my measurements.

The most relevant section of text is in 4.1 as below:

For best performance with any mixer, the impedance/noise match must be optimized. The input impedance of the SA602A is high, typically 3 k in parallel with 3 pF. This is not an easy match from 50 ohms. In each of the examples which follow, an equivalent 50 : 1.5 k match was used. This compromise of noise, loss, and match yielded good results. It can be improved upon. Match to crystal filters require special attention, but are not given focus in this application note

If the input circuit is balanced then the transformer needs to be carefully designed to aim for best noise figure or for best match into the SA602A. Adding turns to the primary will tend to improve the noise figure (in theory) but this will be at the expense of the input VSWR. The Rp of the transformer itself will also play a role here because the transformer will not be lossless when trying to make a transformer that steps up from 50R to several thousand ohms. I think this loading effect from the transformer Rp will tend to bring the noise and power match closer together if the matching transformer has some loss.

There are two pairs of traces in the diagram below. One is for measurements of a real SA602A chip and one is for a basic attempt to design a similar Gilbert cell chip in Genesys that has the same circuit and bias conditions as the SA602A. It's probably best to treat the results for my own Gilbert Cell model as a novelty attempt to copy the SA602A but it did produce the same input impedance and also conversion gain and distortion performance as the SA602A.

[G0HZU_JMR]

For example if you look at figure 9 in AN1993 this is the example circuit they refer to where they choose to match to 1500R rather than the 3000R input Rp of the SA602A.

For best performance with any mixer, the impedance/noise match must be optimized. The input impedance of the SA602A is high, typically 3 k in parallel with 3 pF. This is not an easy match from 50R. In each of the examples which follow, an equivalent 50 : 1.5 k match was used.

I simulated their input network and if you include the 3pF input Cp of the SA602A and maybe 0.5pF Cp for the 280nH matching inductor you can see that their network produces a source impedance that is resonant at 45MHz with their target of 1500R. Yet they clearly state the input impedance of the SA602A is 3k || 3pF. This shows that to get best performance in terms of noise figure it's best to not try for a power match into the chip but to try for their suggested approach where the input network produces a source of 1500R. This gives a good tradeoff between noise figure and input match.

I'm not sure how much lower the noise figure can go if one tried for best noise match. This might mean adding more turns than expected to the primary of the input transformer or (in the case below) changing the capacitive tap to lower the source even lower than 1500R. Usually the best noise match will present a source impedance that is lower than the optimal impedance for best power transfer.

[Alan_G3XAQ]

I find the app note confusing. Are they saying at 4.1 that the input impedance from pin 1 to ground is 3K//3pF with pin 2 is at AC ground (ie single ended) or are they saying the impedance between pins 1 and 2 in a balanced input configuration is 3K//3pF?

Alan

[G0HZU_JMR]

Obviously, when the SA602A is driven with a balanced drive to pins 1 and 2 the transformer also has to be designed to try and find the best compromise between noise figure and power match.

I think I tried 2T and 3T primaries into my 11T +11T secondary when driving the SA602A in balanced mode up at 50MHz. I might have achieved a lower noise figure if I'd added another turn on the primary?

[G0HZU_JMR]

Originally Posted by Alan_G3XAQ

I find the app note confusing. Are they saying at 4.1 that the input impedance from pin 1 to ground is 3K//3pF with pin 2 is at AC ground (ie single ended) or are they saying the impedance between pins 1 and 2 in a balanced input configuration is 3K//3pF?

Alan

I did wonder that but I'm fairly certain they are referring to single ended mode (with pin 2 at AC ground) because the whole of AN1993 uses single ended mode circuits and they state that they opted for a 1500R source rather than trying to match to the 3000R of the input.

I can try measuring the SA602A again with my VNA. I can also do a two port measurement looking into ports 1 and 2 as this VNA has a 'balun' mode.

[Alan_G3XAQ]

This is beginning to make a nonsense of my attempt to terminate the crystal filter preselector in the transformed mixer input impedance!

Alan

[G0HZU_JMR]

I wouldn't worry if your circuit works because this probably doesn't amount to any significant mismatch loss in the filter.

The image below shows my Genesys simulation circuit and it is what I assume is inside the SA602A based on circuits shown in the Philips app notes and some measurements of a real chip and it is also based on a classic Gilbert Cell circuit. Obviously this is a risky thing to do but in order to get the same performance as the real chip I had to bias the base of the BJT at pin 1 (and pin2) with a series 5k resistor as circled in red below.

This means that if you were to measure between pins 1 and 2 with a DMM you would see 5k + 5k = 10k ohm when just measuring the bare unpowered chip between pins 1 and 2.

This resistance is what I see on the SA602A chips I have here. If I measure an SA615 mixer+ limiter chip I see 19k across the mixer input pins. If I try an make a Genesys model of the SA615 mixer with 9.8k bias resistors I get a very similar input impedance as the datasheet for the 605/615. This chip has a slightly higher input impedance and the datasheet suggests about 4.9k in parallel with a couple of pF.

[G0HZU_JMR]

To show the test setup for impedance is OK I measured a 3k3 resistor in parallel with a 3p3 capacitor out to 100MHz.

I then measured the input to the SA602A pin1 with Pin2 AC coupled to ground with 100nF. The PSU was set to just over 5V and the LO port of the SA602A was driven externally from a sig gen at 50MHz.

The VNA power was set to -40dBm and this explains the noisy plots. I also repeated the calibration and measurement with a -55dBm VNA source power and the result was the same although it was getting quite noisy. I have to turn down the VNA power to prevent non-linear effects in the SA602A. It should be fine at both -40dBm and -55dBm RF input levels.

You can see that the 3k3 resistor in parallel with a 3p3 cap measured fine out to 100MHz. the capacitance was slightly higher than 3.3pF because there is probably 0.05pF capacitance in my test fixture and the resistor will probably have 0.1pF self capacitance.

The Cp for the SA602A is a little high at 3.5pF but this includes the short connection wire to the PCB and this will add a small amount of capacitance to free space.

[Radio Wrangler]

Originally Posted by Alan_G3XAQ

This is beginning to make a nonsense of my attempt to terminate the crystal filter preselector in the transformed mixer input impedance!

In general, trying to match an awkward impedance hits you in three places. You wind up with the matching section running at rather high Q, making it very narrowband (not a problem here per se, but troublesome in other cases). This also exaggerates the effects of the losses of your matching components, and it makes the matching circuit place extreme demands on component accuracy and stability.

One part i had to drive as a low level stage in a transmitter needed a matching network that looked like 2.2pF shunt, 2.2pf series, and a few mm of transmission line which acted as the pad for the SMT legs paralleled for the input. This doesn't seem too onerous and I only wanted operation on a single spot frequency.

The sting came in that the part was an HBT in somewhat exotic material, and at lower frequencies its gain went up like a rocket and it had stability problems. It really wanted some driving network which degraded to a dozen Ohms resistive or so at substantially lower frequencies. I couldn't use any complicated network, because the strays then spoiled the match on the operating frequency.

Some parts really are painted into a corner as far as matching is concerned, and good stability can be impossible to achieve.

From a noise figure perspective the first thought is that you want a good match into the input impedance of a part so that you don't waste any of the available signal. If a part has a high real component of input impedance, you want to develop as many volts across it as possible, so the input match looks like a step-up transformation, along with cancellation (exploitation sometimes) of the reactive part. Ditto for current step-up for devices with low real components.

Knowing the input Z of the part, you can now construct a Smith chart with contours of insertion gain for the part showing how it behaves with different Z values presented to its input.

But if you measure the noise figure while you're moving the presented Z around, you get a second contour map. Now for the nasty bit.... Best noise figure rarely happens to be close to the best gain impedance. You have to bite the bullet and try to compromise to get the best overall noise figure, considering you need the gain to dilute the noise contribution of the subsequent stage.

I don't think you need togo to these lengths with an NE602, especially just for checking if there are high levels of QRM but these issues are ones which bite if you do try to optimise something.

The business of varying the impedance presented to a device input or output to explore the effects on gain/noise figure/etc is commonly referred to as load-pull testing.... even when it's the input that's being tested. No wonder people entering the RF field get confused.

David

[Radio Wrangler]

Really stupid sideways thought....

If you only want to measure at a spot frequency, on say the 20m band and you have a crystal preselector filter, why does it need a mixer? You could just have some gain at the intended RF frequency into a detector. You'd need reasonable screening, but you could chuck the NE602 and the LO.

David

[Alan_G3XAQ]

Originally Posted by Alan_G3XAQ

I find the app note confusing. Are they saying at 4.1 that the input impedance from pin 1 to ground is 3K//3pF with pin 2 is at AC ground (ie single ended) or are they saying the impedance between pins 1 and 2 in a balanced input configuration is 3K//3pF?

Alan

Doh, the NE602 input is a differential amplifier between pins 1 and 2 (long tailed pair or similar). So the load is the same whether driven single ended or balanced. In this case it is around 3 or 4K.

A contributor on the qrptech forum got a similar result with a narrow band 14MHz tuned input. 3:14+14 turns presented close to 50 ohms to his N2PK-type VNA running at -40dBm. If the turned transformer is perfect this suggests the NE602 has an input of 4.3K.

With a 3 turn impedance-matched link the loaded Q is quite high: over 8. I'd be uneasy with this if for no other reason than FedEx giving it a good shaking during transport to Africa might detune it. The 5 turn link does seem the best compromise on Q (just over 4), sensitivity, and crystal filter matching (which isn't very sensitive to the mismatch)

Alan

[Alan_G3XAQ]

Originally Posted by Radio Wrangler

If you only want to measure at a spot frequency, on say the 20m band and you have a crystal preselector filter, why does it need a mixer? You could just have some gain at the intended RF frequency into a detector. You'd need reasonable screening, but you could chuck the NE602 and the LO.

David

This was mooted on one of the RSGB forums but I was uneasy about needing around 60dB gain at 14MHz and the need for "reasonable" screening. I think it would need two crystal filters, one at the aerial and one in front of the detector, and probably some gentle filtering along the way as well to keep the broadband noise in check. Using say three MMICs (MAR-6+ or similar) might give the simplest circuit but at 16mA apiece and 8mA for the detector you would be looking at about 60mA current drain, which I suppose would be OK with a PP3 for the short bursts during measurements. The NE602 version draws 20mA.

Perhaps I should try it for a Mk2 design?

73, Alan

[G0HZU_JMR]

I tried measuring the RF inputs with the VNA as a full 2 port network and I did this at a VNA power of -40dBm.

I had to use a DIL packaged device for this as it allowed me to easily connect the fixture directly at the pins of the chip. I used a new (old stock) NE612A for this test and I set the external LO to 50MHz and ran the chip at 5V. The VNA ports are DC blocked so the reference plane is set right at the pins of the NE612A rather than via DC blocking caps.

This produces a 2 port s2p file and this can be post processed on Genesys as in the image below. I've rather lazily done it this way rather than set up the balun mode in the VNA.

I used an ideal transformer to measure it in balanced mode and the results for single ended and balanced mode are shown below. This makes it easy to compare what is seen with a direct single ended measurement and also a single ended view via an ideal transformer.

With single ended mode it shows 3600R || 2.7pF at 14MHz

With the transformer it shows 5700R || 1.7pF at 14MHz

This was measured with a 4 port Agilent E5071 VNA and I used a 4 port N4431B-60006 Ecal module to calibrate the system for a two port measurement and the reference plane was set at the pins of the device. I exploit the fixture simulator feature in the VNA to do this. This system usually works well up to many GHz.

[G0HZU_JMR]

If it helps, here's a simulation of a differential amplifier using a pair of BJT cascodes.

This is very similar to the bottom half of my Gilbert Cell model. The plot below shows the same test and the results are very similar. I measured about 11k ohm across pins 1 and 2 with a DMM at DC so I used 5.5k bias resistors at the base in the model below.

As the frequency is increased above 50MHz, the performance of my model falls away faster than the real NE612AN and this is almost certainly because I've used SMD packaged BJT parts within the model with an Ft of about 2GHz. The NE612AN BJTs will all be together on the same die rather than having connections with all the parasitic issues introduced by SMD packaging.

However, at 14MHz the results in single ended mode and with the transformer are remarkably similar.

[Alan_G3XAQ]

I can't see the circuit clearly in the thumbnail but am I right in believing that in a perfect long tail pair (very large bias resistors, very large beta, zero Cbe,, perfect current source at the emitters, etc) then the input impedance for single ended and balanced drive would be the same? The differences come from the shunting by the bias components. Yes?

[G0HZU_JMR]

Yes, I'd expect a theoretical BJT based diffamp to have the same for single ended and balanced drive but I think this circuit is sufficiently removed from ideal that I think it has to be measured.

[Alan_G3XAQ]

Thanks for that, Jeremy.

I chose balanced input because I thought it might help suppress LO leakage out to the aerial. Now I'm not sure it was worth the effort. I chose a 10KHz audio tail end for the same reason: to put the LO far into the crystal filter stopband. All this was because I've read horror stories about LO leakage causing hum modulation on received signals. The measured LO signal leaking from the input is less than -120dBm.

A single ended input with a 5:28 turn transformer would mismatch my crystal filter a bit less but I'm not going to faff about any more. It's good enough as it is.

73, Alan G3XAQ

[G0HZU_JMR]

The differences come from the shunting by the bias components. Yes?

Yes, I think the bias resistor is quite key in all of this. As I mentioned earlier I think it is possible to estimate the value of this resistor by simply using a DMM across pins 1 and 2 and then dividing by two. I don't think there is a shunt resistor at the base because a DMM check can't see one. I think it is done via a series resistor from an internal voltage reference at about 1.4V. This resistance seems to be between 5000R and 5500R on the 602A and 612A chips I've measured.

On a typical 602A device I see between 10k and 11k ohm between the pins. Therefore I use 5000 ohms as a base bias resistor in my models.

I can only guess at the history of these parts but the old Signetics databooks show the original 602 (602A doesn't exist yet?) as having a genuine 1500R input impedance 'through to 50MHz'

It could be that the early Signetics versions of this chip really did have a 1500R series bias resistor. The early 605 chip has a similar mixer section and Signetics also stated 1500R for this chip in their old databooks.

However, the later 605A shows 5000R and maybe they changed the bias resistors in the A versions of the 602 and 605 chips? On my Genesys model the noise figure improves if this is done.

All this is guesswork though...

[G0HZU_JMR]

I think what spoils the SE vs BAL theory for the input impedance is this 5500R resistor at the base. In single ended mode pin2 is bypassed so this resistor is shorted at AC. However, this 5500R resistor is still there at pin1.

In balanced mode both resistors are effectively in circuit and so this is enough to upset things when the two configurations are compared.

I just tried making the crudest diff amp model using an old s2p model of an MPSH10 BJT. This doesn't need any bias components as it was measured at 3V and 1mA. With just the raw s2p models and no base bias resistor, the circuit does a really good job at demonstrating theory. In the first plot below the input is measured in single ended and also balanced and the result for the input impedance is the same and this agrees with theory.

However, when I add the 5500R bias resistors it upsets things and the plot for this is also shown below. The circuit no longer follows theory for a diff amp.

It's interesting to see that (at low frequencies) even this crude model gives similar input impedance in both states as the real SA602A! Obviously, the MPSH10 is a lower Ft device so the performance suffers.

Note that this is just a pair of MPSH10 devices in a basic diffamp. There is no cascode circuit here and this also spoils the high frequency performance.