The authors have proposed Multi-Input Multi-Output (MIMO)-Constant Envelope Modulation, (MIMO-CEM), as power and complexity efficient alternative to MIMO-OFDM, suitable for wireless backhaul network in which relay nodes are fixed in their positions. One of the major problems to withstand real application of MIMO-CEM is to estimate MIMO channel characteristics. The MIMO-CEM is based upon two contrary schemes; one is the nonlinear CEM modified Maximum Likelihood Sequence Estimator (MLSE) proposed by the authors, which needs accurate channel information to replicate the received signal passing through channel. The other is low resolution analog-to-digital converter (ADC), i.e., 1-bit in the default operation and 2 or 3 bits in the optional operations, which means that the received channel amplitude information may be completely or partially destroyed. These two contrary demands make MIMO-CEM channel estimation a big challenge; how we can accurately estimate the channel in these severe low ADC resolution conditions. We consider this issue through designing an efficient MIMO-CEM channel estimator based upon a block based adaptive filter. In addition, in order to speed up the parameter convergence rate in the proposed adaptive estimator, we design a correlator estimator as initial channel state estimation used in the adaptive estimator. We prove the effectiveness of the proposed MIMO-CEM channel estimation under different channel scenarios and different low ADC resolutions such as 1, 2 and 3-bit.