The downsampling of a discrete wavelet transform (DWT) has a side effect of the lack of shiftinvariance. There are two main solutions for this effect: one is the stationary wavelet transform (SWT), which does not apply downsampling. The other is the complex DWT (CDWT), which uses dual multiresolution analysis(MRA). We choose the CDWT as a target of research. It is well known that wavelet functions become Hilbert transform pairs if the low-pass filters (LPFs) on the reconstruction side have half-sample shifts. In this paper, we propose a quasi-shift-invariant (QSI) CDWT for bi-/orthogonal wavelets as a new CDWT. We report three new works (W1-W3) on it: (W1) we generalized the condition of Hilbert transform pairs and employed a complex wavelet function as a conjugate analytical signal. (W2) We defined a structure that achieves shift-invariance. The structure requires half-sample delays between the inputs of real and imaginary parts. (W3) We proposed an implementation of the QSI-CDWT and confirmed that our method has higher shift-invariance than the conventional CDWT. However, two problems (P1, P2) remain unsolved: (P1) our method requires more resources, such as memory and calculation time, than the conventional CDWT. (P2) Our theory cannot make all packets shift-invariant in a classical wavelet packet transform tree.