I would love to use the capacitor at the input. But, I am sampling 8 channes using 4051 at the highest possible speed.

Ground control to crew of 'Apollo 11' mission, reaching the moon's surface: "Of course, we understand your needs, Mr. Armstrong and Mr. Aldrin. And we really would have loved it to equip the 'eagle' with a rocket motor, but then 'Saturn V' would have been to heavy for lift-off."

Raj,

I have problems to stay cool! Why do you think you are smarter, more intelligent and more experienced than all developement engineers of HARRIS, INTERSIL and MAXIM together? If you can see on EVERY schematic in datasheets this input filter, could it be possible that it should really be used?
When wanting highest precision, during integration phase input signal is not allowed to change too much! Integration phase lasts 10000 periods of clock frequency, means something like 5...10msec. Input filter has time constant of 10msec, according to a corner frequency of 16Hz. So, 50Hz (60Hz) mains frequency and especially their harmonics are highly reduced by this filter. This is essential, because input resistance of ICL7135 is extremely high and tolerable error must be lower than 100µV (!!!).
If you design such a high impedant signal routing, it's nearly impossible to reduce any kind of noise, interference, stray coupling, charge injection, etc., to less than 100µV, WITHOUT any filtering.

Another point: Your circuit is producing so much high frequency noise (think about the clock frequency of about 150kHz (and it's harmonics!) and the parallel interface to mcu), and there's always so much stray coupling and charge injection, that without input filter a totally unaceptable high frequency interference will be present at analog signal input. In combination with unavoidable unlinearity and bandwith limitation of input buffer stage, complex distortion products are fabricated. If you ask me what consequence this can have? Just that, what you experienced, namley introduce of unexpected offset voltage and the impossibility to achieve zero reading!

Input buffer must work with an enormous precision. But precision and speed is a contradiction. So, input buffer circuitry is designed to be precise but slow. But that's no problem, when some input low pass filtering is added. So, input filtering you omitted, is part of ADC circuitry, and you cannot simply remove it.
By the way, high frequeny interference at analog signal input will not only cause trouble during intergration phase, but also during other phases of onversion! And, also unlinearity of internal switches (on-resistance) can introduce unexpected offset voltage, when high frequency interference is not suppressed enough.

There's another point: Don't forget, that CD4051 can contribute much higher leakage current than ICL7135!

Finally: You cannot mutate this very precise and slow ADC into a fast multi channel ADC. This will not work. And you cannot remove essential components of circuitry like input filter. But if you do so, don't be surprised, when big errors occur...


Kai