Memristors have been widely studied as artificial synapses in neuromorphic circuits, due to their functional similarity with biological synapse, low operating power, and high integration density. In this work, a memristor bridge synapse for SNN with excellent linearity and soft-bound synaptic plasticity is designed and utilized for a neuron circuit implementing the robust spike-timing dependent plasticity (STDP) learning. This is the first of its kind demonstrating successful pulse width encoded multiplestep quantized STDP, with mixed-signal neuron possessing linear weight update. Physical models are employed to study the performance of proposed synapse and circuit, and simulations are carried out based on the MATLAB Simulink and Simscape. An improved memristor model which exhibits balanced flexibility, efficiency, convergence, and emulation performance, is developed though including the nonlinear Joule effect and weak signal effect. By using the improved memristor model in pattern learning, the influence of weak signal induced weight variation on circuit performance can be rigorously assessed. Moreover, the robustness and compatibility of the neuron circuit are greatly enhanced by employing the clock-based square-wave pulsed to process and program the synaptic weight. This proposed circuit could give an inspiration for combining the analog memristive synapse and leaky integrate-and-fire neuron with digital control units, prompting their development as edge computing devices.