This article presents propagation of current sensor error through $dq$ -transform and the propagated error effect on high-frequency injection (HFI)-based self-sensing control. Three-phase ac systems use Clark and Park transform with $a, b, c$ phase inputs for control purpose, i.e., $dq$ -transform. When error exists in $a, b, c$ inputs, e.g., due to quantization, gain, offset, noise, and others, the error propagates to $d$ - and $q$ -axes and $dq$ -transform becomes inaccurate; therefore, the control performance degrades. Statistical models based on variance are developed for 2- and 3-channel based $dq$ -transform. The error variance models are verified using the error probability density function (PDF) with uniformly distributed random inputs. It is shown that the error propagated in $dq$ -axes and self-sensing control performance become rotor position-dependent, following the error variance model with error that exists in current inputs. It is shown that the error variance of 2-channel based $dq$ -transform becomes three times higher on average compared to 3-channel based $dq$ -transform. It is demonstrated that 3-channel based self-sensing control results in lesser position estimation error compared to 2-channel based self-sensing with a tradeoff in an additional sensor in the machine drive system.