[Help] LNA & MIXER test in receiver design.

The answer is that people often make test chip with single blocks.
But it also depends on the confidence you have on your design.
In the case of LNA Mixer you can add a buffer between LNA and Mixer, for example, that can be switched off, and test LNA output (be sure that the buffer does not limit the performance of LNA in term of noise, band, nonlinearity), and test the Mixer in the complete LNA Mixer IC. It needs of course more pins.

But if you have a good confidence in your design, you can avoid to test LNA stand alone, and doing accurate measurements, you can understand the single block performances.

I hope it can help.

Mazz

In many cases, you can use your calculations and simulation results together with measured performance of the cascaded LNA-Mixer in order to predict/extract the achieved performance of your design.

All the best

DonJ said:

In many cases, you can use your calculations and simulation results together with measured performance of the cascaded LNA-Mixer in order to predict/extract the achieved performance of your design.

All the best

Click to expand...

You mean from the measurement result of LNA+Mixer, I predict the individual performance, am I correct? But what if the measurement result is quite different from what we expect (simulation)? In that case, I don't think it is easy to find out the problem.

If you are not going to measure it on-wafer, don't put matching network on chip. For a high Q node like output of LNA, it is very hard to get a good power match by using on-chip LC.

I agree Mazz's method, shunt a buffer between LNA output and dedicated test pads. Normally off under LNA+Mixer operation. But a substitution is using two spare metal wires connected to pads and LNA(differential) output. Under normal operation, just keep the pad open. Under LNA test mode, turn Mixer off and match LNA output on board. Under Mixer test mode, turn LNA off and match Mixer input on board.

Layout these wires with top metal layer, if you don't want it anymore, cut it with laser beam.

Thanks a lot !!

dsjomo said:

If you are not going to measure it on-wafer, don't put matching network on chip. For a high Q node like output of LNA, it is very hard to get a good power match by using on-chip LC.

Click to expand...

I totally agree with you, I designed an LNA with on-chip output matching several years ago. The on wafer measurement showed that its output oscillates while my simulation indicated nothing about that. (BTW, the input matching is very good)
I have seen the term "high Q node" several times, but I am not very clear about the actual meaning about that, can you explain?

dsjomo said:

I agree Mazz's method, shunt a buffer between LNA output and dedicated test pads. Normally off under LNA+Mixer operation. But a substitution is using two spare metal wires connected to pads and LNA(differential) output. Under normal operation, just keep the pad open. Under LNA test mode, turn Mixer off and match LNA output on board. Under Mixer test mode, turn LNA off and match Mixer input on board.

Layout these wires with top metal layer, if you don't want it anymore, cut it with laser beam.

Click to expand...

1. what kind of buffer?
2. "on board" means on PCB board, then do the off chip matching?
3. This method seems good, has anyone used this method for testing? I want to following the common and pratical way people do.

Here is a draft illustration.

**broken link removed**

1) The buffer can be a simple differential pair with large bias current driving low value resistors. But you have to know the performance of the buffer in order to derive the actual performance of LNA.

2) Yes, board means your PCB. Mount your die onto PCB, bond, and match the circuit.

3) The method I use can measure both LNA and Mixer but it suffers a serious problem --- ESD. It may cause ESD failure to the input of Mixer if you do not put any ESD device to that test pads.

4) If you really want to know the performance of both LNA and Mixer, connect them outside. Namely, design two irrelevant circuits on the same die. But if oneday you want to connect them, you can connect them on PCB. But the performance will be different compared to direct connected inside the chip. If you are a freshman to your company, do this in the most conservative way for your reputation.

Thanks a lot !!

dsjomo said:

Here is a draft illustration.

h**p://home.pchome.com.tw/mysite/dsjomo/Testkey.bmp

1) The buffer can be a simple differential pair with large bias current driving low value resistors. But you have to know the performance of the buffer in order to derive the actual performance of LNA.

Click to expand...

I got it.

dsjomo said:

3) The method I use can measure both LNA and Mixer but it suffers a serious problem --- ESD. It may cause ESD failure to the input of Mixer if you do not put any ESD device to that test pads.

Click to expand...

Only ESD problem? For my prototype design, I think I can handle this. What's the other problem you encounter?

What is the actual meaning of "high Q node"?

Thnak you.

Yes, there is another severe problem----- the pad position of your LNA output test pads. If the output pads of LNA is close to its input pads, the direct couple might cause K < 1. This couple is resulted from substrate coupling, bondwire coupling and PCB coupling. For a bundled LNA+Mixer, the LNA output is coupled or direct connected to Mixer, so the coupling effect is less considerable.

As we know, bad isolation damages K factor, so one would think that cascode LNA should not encouter this problem. Actually, it can't get rid of the problem, especially for a high gain RF amp like LNA. The S12 value is always largers than -40dB, with bad layout or careless design, S12 may be larger than -25dB.

Whenever you try to pull out the output signal of LNA, be aware of the coupling effect between input and output node. Otherwise you get a oscillator, not LNA.

Added after 8 minutes:

A high Q node means that the impedance attached to that node is mainly inductive or capacitive. If you map the impedance onto smithchart, you'll find the point is located near the circle edge. You can also plot Q line to estimate the Q value on smithchart.

See Microwave Transistor Amplifiers, Gonzalez. Figure 2.4.16, for constant Q line on smithchart.

Other than the ESD and coupling problem, will the long wire and pad affact the performance of the LNA or mixer?

Added after 10 minutes:

I mean other effect like the add on capacitance or ...?

Of cource they will affect your circuit. The Q of bondwire is high so it would not affect the available power gain, but the Q of pad's parasitic capacitance is low because of the substrate resistive parasitics, so it would affect your available power gain.

If you do care the parasitics when the test pads are open and you only want to test the direct connected LNA+Mixer, as i said, cut the metal line with laser beam. While measuring LNA or Mixer, we can match the circuit using on board passive components. And the power gain would be degraded because of the pad's resistive loss.

If you dont like the lossy test pads, another way is to make your test pads ground-shielded, just like how we do to the input pads of LNA.

Anyway, find someone who has the laser beam cutting equipment(that machine is very very very expensive) and draw the critical wires with top metal layer so that if something bad happend, you can cut them off.
P.S. Laser cut might also damage input gate node of Mixer.

Thank you very much dsjomo, you've given me a great help.

1) The buffer can be a simple differential pair with large bias current driving low value resistors. But you have to know the performance of the buffer in order to derive the actual performance of LNA.

Click to expand...

Question about buffer design. I am using a resistive feedback buffer. So what simulations should I run to characterize the performance of the buffer and make sure it does not affect the performance of LNA?

Thanks

ccw27 said:

Question about buffer design. I am using a resistive feedback buffer. So what simulations should I run to characterize the performance of the buffer and make sure it does not affect the performance of LNA?

Thanks

Click to expand...

The simplest way is simulating your LNA both with and without the buffer, then compare the 2 simulation result. But you have to match the output port for the LNA without buffer using ideal L and C component, and ensure that the LNA is unconditionally stable.

If the LNA core is not unconditionally stable, it is impossible to match the output port. Some other simulation technique should be taken.

Thanks.

So we should first match the LNA output using ideal passive device and check its S-parameters and so on. Once this is done we should attach the buffer, match the output and check the performance. Do I need to match the output of the buffer? My question is buffer is bound to increase NF, affect gain. Though you can roughly estimate the perfomance of buffer in simulation,its bound to be different in real silicon. So once the chip comes back and we measure the LNA+buffer we might be either over or under estimating its performance. Is that true?

Thanks

ccw27 said:

Thanks.

So we should first match the LNA output using ideal passive device and check its S-parameters and so on. Once this is done we should attach the buffer, match the output and check the performance. Do I need to match the output of the buffer? My question is buffer is bound to increase NF, affect gain. Though you can roughly estimate the perfomance of buffer in simulation,its bound to be different in real silicon. So once the chip comes back and we measure the LNA+buffer we might be either over or under estimating its performance. Is that true?

Thanks

Click to expand...

Yes, your chip is hard to be exactly the same to the simulation result.

There are several methods to measure the performance of LNA which is followed by a mixer indirectly.

(1) Using buffer. And the buffer should present a similar load impedance to the LNA core compared to the input impedance of mixer. Under this situation, the Zload of the LNA core is similar to the real case. But the buffer would damage the overall performance, so we have to minimize this effect. By the way, it is impossible to make an additional buffer testkey, so the only practical way might be estimating the error introduced by buffer under simulation.

The output port of buffer should be power matched. You can use resistors as loads, and of course it would generate noise.

So this method should give you a reliable measurement result on gain, especially for the case when LNA+mixer directly to estimate the volatage gain. If the buffer is linear enough, the linearity measurement is also reliable.

(2) Directly measure the LNA core. This method is to match both ports of LNA core then measure. So if the LNA core is unconditionally stable, its performance should be in the best situation. But, the Zload is different to the Zin of mixer. So the measurement result only represent for the maximum achievable performance of your LNA core, and if you connect the LNA core with mixer, the performance would certainly be worsened.

This method should gives you a reliable NF measurement result. Since there are no extra ciruits following the LNA core(except the parasitics of output port matching components).

(3) Directly measure LNA+mixer.

To directly measure LNA+mixer, you also need a buffer after the LNA right? How should you design the buffer?

Thanks

Common Drain -> better linearity
Common Source -> better noise figure
Common Gate -> better isolation

How should I design the buffer after the Mixer? I need to convert from differential to single ended output. For differential I can simply use two source followers but how about differential to singled ended? Please suggest a topology.

Thanks

According to my calculations using a source follower as a buffer for mixer output seems to introduce a lot of noise since its gain is less than unity. But since it is the third stage of a cascade stage it may still be ok. Should I use a source follower or a common source as a buffer?

Thanks