How can I keep two channels gain difference constant?

I have two same configuration receiver channels, the input power is 3dB power splitting and then input to the two channels respectively.
The gain difference is constant when input power is between -30dBm and -80dBm. And when input power is between -81dBm and -113dBm, the gain difference of the two channels is deviated from the constant.
How to keep the gain difference constant when input power at -90dBm ~ -110dBm level?
I know there are many Electrol-magnetic interferences at environment. So what should I do? Replace the normal RF cable with more shielded cable? Any others factors?

Hi Tony,

it seeams that what you are measuring is not really gain but the ratio (output_signal + output_noise) / input_signal, that deviates from gain when signals are weak.
Look at the level you measure at the outputs when inpul signal is 0 (input terminated with a matched load). You will have an idea of the output_noise term in the above expression.
Regards

Z

Hi, zorro,
Every time at the beginning of the test, out system samples the noise floor for each of the two channels.
And the sampled data was averaged by more than 2000 times. Our noise floor ia about -80dBm at the receiver output.
The noise figure is less than 4dB from the front-end of the LNA to the back-end of ADC, and gain is about 33dB.
The sensitivity is about -113dBm when SNR=3dB.
But when SNR<=15dB, the two channels gain difference deviates the constant more than 0.1dB.
So how to solve this problem?

Hi Tony,

A gain difference of only 0.1 dB could be comparable with or less to the measurement error.

That difference (or a part if it) can be also originated in small differences between the two channels. I'll try to explain with this example:
Imagine for instance that you have two channels with identical gains but with a difference in their noise floors of 1.5 dB.
When the signal is strong, noises are negligible and you see the same output level in both channels.
Now suppose you decrease the signal in such a way that SNR is 10 dB in the first channel. Signal power is 1 and noise power is 0.1 (in some linear power units, not dB) at the output of this first channel. In the second channel, signal power is 1 but the noise level is about 0.07 (1.5 dB below 0.1, that is the noise in the first channel). You "see" a difference between channels of 10*log10(1.1/1.07)=0.12 dB.
Can be this case similar to yours?

Regards

Z

Hi, ZORRO,

thank you very much.