PCI Classification in 5G-NR: Deep Learning Unravels Synchronization Signal Blocks

Accurate detection of the Physical Cell Identity (PCI) is critical for rapid synchronization and connection establishment in 5G New Radio (5G-NR) systems. This paper introduces a deep learning-based approach for PCI classification, aiming to mitigate the computational complexity associated with traditional methods that rely on decoding the Synchronization Signal Block (SSB). Our approach processes only time-domain baseband samples of the downlink signal, arranged in fixed-length windows. These windows are inputted into a pre-trained Convolutional Neural Network (CNN), which classifies the samples into one of several known PCI values (representing nearby cells) or into an ‘other’ category (representing all non-nearby cells, as well as windows that do not contain SSB samples). Because PCI-related information is contained only in the SSB symbols of a frame, it is possible for an input window to include no or a few SSB samples. Accordingly, in labeling the training set, we use a threshold <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{T}_{\text{train}}$</tex> on the fraction of the samples within a window: if <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{T}_{\text{train}}$</tex> or more of the samples belong to an SSB of cell with a target PCI value, the true label for that window is set to that PCI value; otherwise, it is set to ‘other.’ A separate threshold <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{T}_{\text{test}}$</tex> is used for labeling the test windows. We also study another labeling mechanism whereby only samples of the third OFDM symbol in an SSB (which contains the Secondary Synchronization Signal) is used to determine the label. Our analysis considers two commonly used SSB formats that correspond to 15 and 30 kHz subcarrier spacings, respectively. Extensive simulations are conducted which reveal that the proposed classifier can reliably (above 98%) identify the PCI value of a captured signal even under Signal-to-Noise Ratio (SNR) values as low as -10 dB. This performance comes with a significant reduction in computational complexity as it bypasses the need for traditional SSB decoding procedures used for PCI estimation in 5G networks.