978-1-4244-4547-9/09/$26.00 ©2009 IEEE TENCON 2009 Bi-Directional Timing Recovery for Magnetic Recording Systems Chanon Warisarn and Pornchai Supnithi Faculty of Engineering and I/UCRC in Data Storage Technology and Applications, King Mongkut’s Institute of Technology Ladkrabang, Bangkok, Thailand Email: s9060053@kmitl.ac.th Piya Kovintavewat Data Storage Technology Research Unit Nakhon Pathom Rajabhat University, Nakhon Pathom, Thailand Email: piya@npru.ac.th Abstract—Magnetic recording systems employ conventional timing recovery to synchronize the sampler with the readback signal. However, conventional timing recovery does not perform well when the timing error is large. This paper proposes the bi- directional timing recovery, which utilizes conventional timing recovery to sample the readback signal both in forward direction and in backward direction. The outputs of these two operations are averaged and sent them to the Viterbi detector to determine the most likely input sequence. Results indicate that the bi- directional timing recovery performs better than conventional timing recovery, especially when the timing error is large. Keywords- Bi-directional timing recovery; conventional timing recovery; perpendicular recording; timing error. I. INTRODUCTION Timing recovery is the process of synchronizing the sampler with the received analog signal. Sampling at the wrong times can have a devastating impact on overall system performance. Therefore, the quality of synchronization is very important for all applications. Practically, magnetic recording systems employ the conventional timing recovery with a 2nd-order phase-locked loop (PLL), which consists of a timing error detector (TED), a loop filter, and a voltage controlled oscillator (VCO), as illustrated in Fig. 1. Many timing recovery systems have been proposed in the literature [1], [2], [3]. Most of them can be categorized into two types, namely deductive timing recovery and inductive timing recovery, depending on where the timing information embedded in the received analog signal is extracted [1]. Specifically, the deductive (or feed-forward) timing recovery extracts the timing information before the sampler, whereas the inductive (or feedback) one extracts the timing information after the sampler. However, both timing recovery architectures utilize a PLL to find the location to sample the received signal. Because the inductive timing recovery is widely used in many applications [1], it will then be referred to as conventional timing recovery, whose architecture is shown in Fig. 1. This paper proposes a simple timing recovery architecture, which consists of two timing recovery blocks running in parallel as depicted in Fig. 2. The first block (i.e., branch A) employs a conventional timing recovery to sample the readback signal, while the second block (i.e., branch B) reverses the whole readback signal before passing the reversed readback signal to conventional timing recovery. The outputs of the two timing recovery blocks are averaged and sent the resulting sequence to the Viterbi detector (VD) to determine the most likely input sequence. We refer to the proposed timing recovery architecture as “bi-directional timing recovery.” It can be seen in simulations that the bi-directional timing recovery can help improve the system performance if compared to conventional timing recovery. This paper is organized as follows. Section II describes our channel model and explains how conventional timing recovery works. The bi-directional timing recovery scheme is described in Section III, and its performance is compared with conventional timing recovery in Section IV. Finally, Section V concludes this paper. II. SYSTEM DESCRIPTIONS We consider the perfectly equalized PR-II channel model shown in Fig. 2, where the readback signal can be written as () ( ) () 1 0 L k k k s t aht kT nt τ - = = - - + ∑ , (1) where k a ∈ { } 1 ± is an input data sequence of length L with bit period T , () ht = () 2 ( ) ( 2 ) p t pt T pt T + - + - is a PR-II pulse, () sin( / ) /( / ) p t tT tT π π = is an ideal zero-excess-bandwidth Nyquist pulse, and () nt is additive white Gaussian noise (AWGN) with two-sided power spectral density N 0 /2. The timing offset, k τ is modeled as a random walk model [4] according to 2 1 (0, ) k k w N τ τ σ + = + , (2) VCO A/D Loop filter TED Timing error estimate ˆ k ε Received signal () y t Sampling phase offset ˆ k τ k y ˆ k r Symbol detector Data detection Figure 1. A conventional timing recovery system.