BD-13 An MLP-based ITI Suppression Method for Multi-Head Multi-Track Bit-Patterned Magnetic Recording

Bit-patterned magnetic recording (BPMR) is a promising technology that can achieve an areal density (AD) of hard disk drives up to 5 Tb/in2 [1]. However, it faces significant intertrack interference (ITI) issues at high ADs, which can degrade system performance. Various methods have been proposed to address ITI. For instance, Navabi [2] proposed a 2-dimensional (2D) equalizer and a 1D target for a 3-head 1-track (3H1T) BPMR system. This approach decodes only the main track and outperforms the system using a 1D equalizer and a 2D target. Karakulak [3] introduced the 2D Viterbi detector for the 2D target to counteract the ITI effect. Fig. 1 shows the 3H3T BPMR channel model with the proposed multilayer perceptron (MLP) -based ITI suppression method. An input sequence of the j-th track and the k-th bit aj,k ∈ {±1} is sent to the channel matrix H [1] and corrupted by additive white Gaussian noise {nj,k} to obtain a readback signal {yj,k}, where j ∈ {±1, ±2, 0}. We assume the ITI in the readback signal is caused by only two adjacent tracks. Then, the readback signals {y-1,k, y0,k, and y1,k} are fed to an MLP-based estimator to approximate the ITI-unaffected signal of the main track {r0,k}. This ITI suppression method uses an MLP with 27 input nodes and 5 hidden layers, each containing 100 nodes. All hidden nodes employ a leaky ReLU activation function, and the output node uses a linear activation function. The signal {r0,k} is scaled by a weighting factor w and subtracted from {y-1,k} and {y1,k} to lessen the ITI effect caused by the main track [4]. Hence, the refined signals {r-1,k} and {r1,k} will contain only the one-side ITI effect, which will be equalized by a 1D equalizer designed for the 2×3 2D target [4]. Conversely, the signal {r0,k} can be equalized by the 1D equalizer designed for the 1D target. Finally, the equalized signals {r-1,k, r1,k} and {r0,k} are passed to the 2D and 1D Viterbi detectors to determine the most likely input sequences {a-1,k, a1,k} and {a0,k}, respectively. Fig. 2 compares the system performance at AD = 3 Tb/in2. The proposed system performs better than the conventional systems presented in [2] “Conv 3H1T” and in [4] “Conv 3H3T.”References: [1] Y. Shiroishi, et al., “Future options for HDD storage,” IEEE Trans. Magn., vol. 45, no. 10, pp. 3816–3822, October 2009. [2] S. Nabavi, “Signal processing for bit-patterned media channel with intertrack interference,” Ph.D. dissertation, Dept. Elect. Eng. Comp. Sci., Carnegie Mellon University, Pittsburgh, PA, 2008. [3] S. Karakulak, “From channel modeling to signal processing for bit-patterned media recording,” Ph.D. dissertation, Department of Electrical Engineering, University of California, San Diego, 2010. [4] S. Koonkarnkhai and P. Kovintavewat, “An iterative ITI cancellation method for multi-head multi-track bit-patterned magnetic recording systems,” Digital Communication and Network, vol. 7, no. 1, pp. 96 - 101, 2021.