In modern communication systems, a sample rate
conversion is necessary since often the system clock is fixed at
some specific rate. Such resampling is critical because there exists
a tight coupling between the data rates and sampling rates. It
is desirable to have a flexible, high performance, and resource
efficient resampler that can accommodate various required data
rates. To achieve these objectives, we present a novel multirate
resampling method based on graphics processing units (GPUs).
We also extend this method to multiple dimensions, which
enables efficient processing of multiple channels across different
frequency bands. The overall result is a flexible, high-throughput,
and low-latency implementation that is capable of processing
many channels simultaneously.