Filter Design - no response on the GPB

[strabush]

Filter impedance

The filter frequency response depends on the impedance it sees. If you designed the filter to be loaded by 4K at both sides and then you connect it to a circuit that shows it 400 Ohm at one side, it will not work as intended and its frequency response will be distorted. You can put a 3.3K resistor in series of the filter input in order for the filter and RFIC to see the right impedances, but then you will have a 5 - 6 dB loss at the passband. Is the load of the filter 4K too? What RFIC are you using?

 

[brmadhukar]

Hi,
The RFIC is MAXIM MAX2740. The second IF out is at 15.42MHz. The output impedance of the device is supposed to be around 400 ohms but, the tech help told me to use 4K as input impedance in order to provide higher impedance. The output is 4K impedance, input impedance of the next stage amplifier. And my requirement of 10MHz BW at 15MHz is definitely not helping much.
brmadhukar

 

[brmadhukar]

Hi,
I am using the HPF + LPF combination now. I am able to get a decent spectrum, but I am skeptical about the phase response. I cannot measure the phase response ( I have only Spectrum ANalyzer). How do I verify the same.
brmadhukar

 

[strabush]

Phase response

If your filter is a cascade connection of ladder topology low pass filter and ladder topology high pass filter then there is a unique mathematical connection between the amplitude frequency response and the phase response (minimum phase filter).
The easiest way to asses the phase response is to simulate the filter using a simulation software (SPICE, GENESYS, Etc.) and tweak the component values to get the same amplitude response as measured (it is enough to match the responses to the -20dB points). Then the analysis phase response should be identical to the real implementation phase response.

 

[brmadhukar]

Hi Starbush,
I actually use a differential ckt by halving the impedances of series elements and creating one more series path. Will this still be consistant with the phase response from the single ended simulation.
brmadhukar

 

[strabush]

Balanced circuit

If what you do is create a balanced circuit with differential signal passing through it then it still is a one path filter, and what I said is still valid.

 

[brmadhukar]

Hi,
I know the problem with my filter is high impedance. I had asked the Maxim tech help on how about lowering the impedance requirements. (Actually the output impedance of the mixer is around 100-400 ohms but an impedance of 4K is recomended.) I got the following answer.

"As the filter impedance is lowered, the mixer output will see the lower
impedance. This would tend to degrade the mixer's IP3, and 1dB compression point."

Does it mean that we cannot achieve matching impedances without sacrificing something. Can it be a case with this device only.

 

[webdog]

Hi,
I know the problem with my filter is high impedance. I had asked the Maxim tech help on how about lowering the impedance requirements. (Actually the output impedance of the mixer is around 100-400 ohms but an impedance of 4K is recomended.) I got the following answer.

"As the filter impedance is lowered, the mixer output will see the lower
impedance. This would tend to degrade the mixer's IP3, and 1dB compression point."

Does it mean that we cannot achieve matching impedances without sacrificing something. Can it be a case with this device only.

In short: Yes! There is no free lunch here.... if your mixer/amp have 400 ohm output your filter must see 400 ohm or have strange passband/stopband.

Why not go half way, put in a series resistor say 1500 ohm and design your filter (may it be bandpass or lpf/hpf) for a 2000 ohm impedance at the input. I am not sure what impedance you have on the input after the filter, but you should have this port matched too. Try to avoid series resistors as you will have too much losses otherwise.

If the design cannot stand any losses, why not look at high-frequency op-amps? Both Analog Devices and Ti/BB does have suitable amps that will give good match and high performance! You might even end up designing the filter for 50 ohms this way, enabling you to test the circuit with your spectrum analyzer.

Cheers
WD

 

[toonafishy]

Stop thinking like an RF engineeer for a minute and think like an IC designer. Every circuit in the world does not have to be conjugately matched in order to achieve maximum power transfer. In this case, as you apply a lower and lower impedance to the mixer, more and more power is drawn out of the mixer. If the impedance presented is much lower than 4K, then the mixer devlivers too much power and it will distort.
When you plug a power supply into an AC outlet is the power supply conjugately matched to the power grid? If it was, you would burn your house down!

 

[flatulent]

intercept point

At this point in your receiver the noise figure has been set by the RF amplifier and there are still a wide range of signals coming through (This is why the filter is there to eliminate all but one of them). IMD intercept point is the most important characteristic when selecting the load for the mixer. You should follow the manufacturer's advice and use the load impedance they say gives the highest intercept point.

Since the filter will be nearly zero or infinite impedance out of band, you should follow the mixer with a buffer stage so that the mixer sees the proper impedance at all frequencies. Another way is to parallel your filter input with a high pass filter driving a load of the same impedance. That way the mixer sees the proper load impedance at all frequencies. This trick is called a diplexer.

 

[brmadhukar]

Hi,
How do I debug my filter. I have designed a BPF of 10M bandwidth at 15MHz. The simulation is OK, but when implemented I have the following problem.
1) There is no spectrum after 15MHz. I am using cheby I, 0.1dB ripple.
2) No of peaks in spectrum is 3, as no of reactive elements is 8, of 2 series and two parallel resonators. The spectrum seems to have shrunk.

 

[flatulent]

some methods

Is the measured ripple the same as desired? If larger, this usually means that your source and load impedance is off.

Measure the parts on an impedance bridge to see their parasitic resistance and self resonant frequency. Put these values in your simulator.

Are you including the PC trace shunt capacitance in your design?

Wire the filter together in the air like a rats nest by soldering part leads directly together. This will minimize PC trace capacitance effeccts.

Make sure the inductors are not parallel and close together, This will cause mutual inductance.

 

[webdog]

Stop thinking like an RF engineeer for a minute and think like an IC designer. Every circuit in the world does not have to be conjugately matched

You're probably correct, but there are a few points where I disagree.

The filter must be terminated in a matched impedance or the passband will be distorted, the mixer on the other hand can drive most high impedance levels with ease. As said previous in this thread one way of acheiving this match would be using terminating resistors which in this case will degrade the s/n. If the gain is so high that a loss of 10 dB would not matter, it is my beleif that it is far too high gain prior to the IF filter and thus leading to poor IMD.

By using an op-amp after the mixer we can optimally terminate both the filter AND mixer. As a bonus it allows us to design the filter with an arbitrary source/load impedance.

BTW what is the post-mixer amplifier (IF) input impedance?
Cheers
WD

 

[brmadhukar]

The mixer output impedance is 100 ohms. I am now using 1K-1K filter by providing 900 ohms at the beginning of the filter and using 1.3K at the output as 1.3*4/(4+1.3) == 1

 

[webdog]

The mixer output impedance is 100 ohms. I am now using 1K-1K filter by providing 900 ohms at the beginning of the filter and using 1.3K at the output as 1.3*4/(4+1.3) == 1

I just ran a simulation using the source with 100 ohm and a load with 4000 ohm. The filter was a bandpass 3 pole with 1kohm i/o impedance. Using the 900 ohm input filter and the 1.3 kohm load you will have 16 db loss ompared to a perfect match without resistors. This might be tolerated?. 10 dB loss at the input and 6 dB at the output.

With this simple filter the passband curve shows very little attenutation on the high side but fair performance on the lower side. Maybe you could get away with just a low-pass filter? It depends on where your unwanted signals appears.


Good Luck and keep us posted on your progress

/WD