The increasing demand for wireless communications systems had led to the development of advanced multiple-access schemes such as Direct-Sequence Code Division Multiple Access (DS-CDMA). Some challenges faced by DS-CDMA systems are multi-user interference (MUI) and multipath propagation. Decision feedback detectors can provide nearly optimal performance in the presence of both MUI and multipath, but require initialization using training data, which reduces the system throughput. Minimum Output Energy (MOE) detectors, which are based on linearly constrained optimization, can operate "blindly", i.e. they do not require training data, and are effective at suppressing MUI. However, they are sensitive to mismatch caused by multipath propagation, and do not achieve the level of performance of decision feedback detectors. We present a blind detection technique which combines decision feedback with linear and quadratically constrained optimization. The technique outperforms MOE and robust MOE detectors, and achieves a performance level very close to decision feedback. The detector is implemented using recursive least squares (RLS) adaptation and also converges faster than the other techniques.