Proposed is a systematic structure for producing quasi-orthogonal space-time block codes (QO-STBCs) to achieve delay optimality for any even number of transmit antennas. These codes are simple extensions of the conventional ABBA code with four transmit antennas, but they are based on a different symbol pairing that uses an independent cyclic-shifting operation for each input sub-vector. When compared to the conventional QO-STBCs, the proposed codes have significant benefits for receive decoding complexity, or in their averaged error performances.Introduction: Quasi-orthogonal space-time block codes (QO-STBCs) [1-4] have been greatly preferable since an independent joint maximum-likelihood (ML) detection of two complex [1 -3] or four real symbols [4] is available at a receiver. However to reduce decoding complexity, it is more critical to minimise codeword length. This implies that under a condition of full diversity, the time length of a codeword matrix should be identical to the number of transmit antennas, which is called as delay-optimal [5]. However, the existing QO-STBCs [2-4] are delay-optimal only when N = 2 n transmit antennas. For other antennas, the codes are constructed by deleting certain columns from these delay-optimal codes. Therefore, considering the ML decoding complexity and data latency at the receiver, it is meaningful to design a codeword matrix of square size.So, this Letter proposes new QO-STBCs that satisfy the delay optimality with any even number of transmit antennas. These codes are designed based on the conventional ABBA code [1] using four transmit antennas. To enjoy full diversity gain, a constellation rotating method is adopted. Product diversity is also maximised by optimising a rotation angle through a computer search method.