With the evolution of access systems by network function virtualization (NFV) and the software-defined network (SDN), further softwarization including transmission functions is required in order to fully utilize the advantages that softwarization offers to network operators in terms of more flexible and rapid service creation and migration. The softwarization of digital signal processing (DSP) is a prime target, but frequency offset compensation (FOC) has serial processing, which makes its softwarization difficult. In particular, the pre-decision-based angle differential estimator (PADE) of wide-range FOC needs long serial iterations since it obtains stable solutions by using small loop gain in the loop filter. This paper proposes a multiplication-based FOC implementation and a novel multi-thread PADE algorithm for real-time softwarization. The multiplication-based implementation enables parallel compensation of phase rotation by multiplying the estimated CFOs by sampling numbers for Viterbi-and-Viterbi (VV) method and PADE. In addition, our multi-thread PADE algorithm significantly reduces the iterations needed for FOC estimation; using multi-threading, it estimates multiple frequency offsets using the loop filters with large loop gain. This calculation needs relatively few iterations, but its outputs are very noisy, and final frequency offset estimates are obtained after removing the noise by clustering and averaging the results. The results of simulations and experiments, which use a general-purpose server with a graphic processing unit, show that our PADE algorithm is 1,308 times faster than the conventional implementation and 5-Gb/s real-time processing is achieved while maintaining the FOC performance.
INDEX TERMS SDN, PON, Softwarization, DSP and frequency offset compensation
I. INTRODUCTION