I'm sorry, but none of you has quite got it right - you've all picked up on certain points of detail , but you haven't understood the problem and why it occurs.This is not an easy subject to explain in a couple of paragraphs on a website like this but, in broad terms, Schuler tuning works like this:
You have to keep the INS accelerometers level with respect to Earth gravity because, if you don't, they will pick up a small component of gravity and misinterpret this as lateral acceleration. You can't use gravity to reference the vertical and keep the platform level because you're trying to use it to measure accelerations - so you can't use as a datum the thing that you are trying to measure.
So you have to keep the accelerometers level to Earth gravity by using gyros. The trouble is, gyros are rigid with respect to inertial space, not to an Earth-based frame of reference. As you travel over the Earth, the direction between you and the centre of the Earth (ie, the gravity vector) changes. So you have to tell the gyro this. It's called the vertical component of Transport Wander(or Profile Rate, in some books). The way you correct the platform/gyro is to have a feedback circuit from the velocity integrators which keeps the gyros level to Earth. It's called the V/R loop (for latitude) or the U/R loop (for longitude).
Incorporating a V/R feedback loop into the design of an INS/IRS to keep the platform level is called 'Schuler tuning' - a fancy buzzword which, in my opinion, tends to confuse.
However, if the initial conditions leading to the calculation of the velocity are in any way in error (which they are on nearly every flight, to a very small extent), then the V/R loop oscillates - at a period of 84.4 minutes, giving you a position error (usually up to a couple of miles or less) which will never increase to beyond its maximum amplitude, and then repeats itself in a perfectly sinusoidal pattern every 84.4 minutes.
Schuler Tuning is used in both the INS and in the IRS and so you get Schuler period oscillations in both the INS and IRS.
It is not possible to remove it within the INS/IRS, but if the output of your IRS goes to an FMS, the error is mathematically modelled and some of it is removed within the FMC.
Hope this helps.