Abstract-This work focuses on a two-way denoise-andforward relaying system using non-coherent Differential Binary Phase-Shift Keying (DBPSK) modulation, which has the welldefined relay denoising function when channel state information is unknown. We first design the relay denoising function and source decoders using Maximum Likelihood (ML) principles for the general case with parallel relays. As the ML denoising function is hard to manipulate, we approximate it as a multi-user detector followed by a physical layer network coding encoder and obtain the closed-form relay decoding error. For the singlerelay case, we show that the ML source decoder is actually equivalent to the typical DBPSK decoder for the relay-source channel and thus derive the exact end-to-end Bit Error Rate (BER). To minimize the average BER, we also investigate the power allocation problem by use of asymptotic analysis at high Signal-to-Noise Ratio (SNR). We show that the optimal source power is inversely proportional to the square root of the channel gain of the source-relay channel, and the optimal relay power decreases with SNR. For the multi-relay case, though the exact analysis is intractable, we develop upper bound and lower bound on BER and show that the diversity order is exactly ⌈ 2 ⌉ .Index Terms-Two-way relaying, differential modulation, BER, diversity order, power allocation.