Basic analog signal processing question

I'd like to perform an addition of differential current signals in order to feed the resulting differential current signal into a gilbert-based multiplier. How can I perform this addition? Simple KCL won't work on differential signals because the common mode needs to be cancelled.
Thanks for any suggestions.

I could not catch what you meant here. But in case of the removal of the common mode signal, why won't you use ac coupling?

Isn't AC coupling for voltage signals?
Anyway, I am talking about current-mode circuits. It turns out that if I can keep the common mode current within tolerable limits, I won't need to cancel it, although it would be nice to be able to.

I am not familiar with current-mode circuits, but I think there should also be some ways for current ac-coupling. AC coupling is basically a high-pass filter. For example, a resistor paralleled to ground, followed by a series cap to load, might work. The dc current will go to the ground through the resistor, while the ac will pass through the cap if the cut-off frequency is carefully picked up. Correct me if I am wrong.

You can convert from differential current to voltage and then do the combining of the voltages in the usual op amp way.

To do the differential current conversion, use the traditional four resistor-opamp circuit and do not use the two resistors that go to the source of the signal. Use just wires. The current-voltage conversion factor is then the equal value resistors, one from the output to the inverting input and one from the noninverting input to ground.

Another crude method is to put equal value resistors from the two current inputs to ground. This will produce two voltages. From these two voltages go through a differential amplifier.

Then there is the possibility of using the differential inputs of the mixer directly on the two resistors described in the previous paragraph. If your mixer has enough common mode dynamic range, it will handle the input signal properly.

Thanks for your ideas!
Converting to voltage would create extra overhead in my circuits and probably reduce my dynamic range. I wish to maintain a current mode design where I could.
I have found a reference that shows a current mode solution and I will try it. It's in a 1995 paper by Coggins 'A Hybrid Analog and Digital VLSI Neural Network for Intracardiac Morphology Classification'.