We propose a coupled transmission line (CTL)-based load network structure for the Doherty power amplifier (DPA) with an extended output power back-off (OBO) range, designed to achieve high efficiency and broad bandwidth. This load network features a highly simplified structure, consisting of a CTL and a post-matching network (PMN) with only two additional shunt elements. We also propose a novel approach to comprehensively analyze the CTL using its equivalent circuit model, which facilitates the comparison of various CTL-based load network structures suitable for DPA applications. By searching for 2-D closed-form parameters derived from the equivalent circuit models, we can determine the most suitable load network structure and its design parameters for optimal performance across a wide bandwidth. For implementation, a new method is proposed to ascertain the initial dimensions of a CTL based on analyzing the half-circuits as either a coplanar waveguide (CPW) or a CPW with a lower ground plane (CPWG), under even-mode and odd-mode conditions. To validate the proposed CTL-based load network and its equivalent circuit model, a DPA featuring an asymmetric structure using GaN-high electron mobility transistors (HEMTs) was designed and implemented, extending the back-off range to 8 dB for the broad frequency band, including the sub-6 5G (N78) band. The chosen load network, including the CTL, was fabricated on a two-layer printed circuit board (PCB). Measurement results show a drain efficiency (DE) of 44%–52.3% at an OBO of 8.0 dB from the saturated output power of approximately 43 dBm across the frequency range of 2.9–4.1 GHz under continuous wave (CW) signal excitation. With a 5G new radio (NR) signal featuring a bandwidth of 100 MHz and a peak-to-average power ratio (PAPR) of 7.8 dB, efficiencies of 48.0%–55.3% were achieved at an output power of 35.5 dBm, while maintaining an adjacent channel leakage power ratio (ACLR) of under −47.0 dBc across the frequency band after linearization using a digital pre-distortion (DPD) based on a memory polynomial.