T-SPP: Improving GNSS Single-Point Positioning Performance Using Transformer-Based Correction

Algorithm 1

D-Tran model training.

	Input: input data S, learning rate, hyper-parameters
	Output: a well-trained D-Tran model.
(1)	initializing parameters of the D-Tran model
(2)	normalize S
(3)	calculate
(4)	calculate
(5)	for training episode  = 1, N:
(6)	predict and with the inputs of S
(7)	calculate loss based on and
(8)	calculate gradient
(9)	update parameters of DAE
(10)	update parameters of Transformer
(11)	end for
(12)	return the well-trained D-Tran model