A Crystal-Less Frequency-Modulation Transmitter IC with Joint Neural-Network-Driven Modulation and Coding for Low-Power Connectivity

As the Internet of Things (IoT) expands, low-power wireless communication solutions remain critical for applications like smart-home automation and wearable health monitors. While IoT gateways can provide high-performance RF transceivers and powerful digital processors for a robust downlink, IoT devices perform under stringent power/energy constraints that often limit the reliability and range of the uplink (Fig. 26.3.1, top). In existing solutions, non-coherent frequency-modulation schemes such as BLE and LoRa have been widely adopted in IoT nodes to simplify design, reduce cost, and lower power consumption. Traditionally, such transmitters employ direct frequency modulation via a PLL [1]–[6] for stable carrier frequency and low phase noise (Fig. 26.3.1 mid). This simplifies the design by removing quadrature I/O modulation and mixers. However, this approach still incurs the high-power consumption and complexity of the PLL, and the cost and power overhead of the crystal frequency reference. To address this, crystal-less radios using a free-running LC oscillator have been proposed [7]–[9] with simpler and lower-power transmitter (TX) architectures. However, these suffer from carrier drift, excessive phase noise, or requiring additional resonant devices, limiting their use cases. To overcome this issue, [10], [11] calibrate the L0 to the received signals during runtime, at the cost of TX-complexity and power overhead, offsetting some of the power gains from eliminating the crystal reference.