一般而言，GNSS載波相位定位之精度較電碼定位高，利用載波相位觀測量進行衛星測量求解位置時，如何快速得到正確的整數相位模稜值，是求解精度與效率的關鍵。但是參數間彼此高相關，會使這個目標變得困難。這個問題能夠藉由衛星幾何的改變來改善，但是卻因此而加長觀測時間。因此LLL技術以及白化濾波即將原在高相關空間的參數，投影至另一低相關空間。使數學變換過程等效於衛星幾何改變，進而可在較短觀測時段裡求解。 LLL是將一正定對稱矩陣分解為上下三角矩陣，再利用Gram–Schmidt正交變換將三角矩陣之向量轉變為彼此間皆正交，接著藉由正交化後之上下三角矩陣相乘得到一具備對角優勢之協方差矩陣。 白化濾波是利用Crout 因子分解，使一正定對稱矩陣分解為對角線矩陣與單位上下三角矩陣之連乘。應用其矩陣對角線化的特性於相位模稜實數解的協方差矩陣上，產生具備對角優勢之協方差矩陣。 應用此協方差矩陣可大幅減少整數相位模稜的候選解。最後將候選解逐一代入觀測式中重新進行平差演算，求取一殘差二次形為最小之解。 Generally, the GNSS carrier-phase is more accurate then the pseudorange. While using carrier-phase for positioning, the key point is how to obtain the correct integer ambiguity quickly and efficiently. However the high correlation between parameters makes it to be difficult. The problem can be improved by the changing of the geometric of satellites. But it needs longer observation time to reach. Therefore the LLL algorithm and the whitening filter are techniques mapping the parameters from a higher correlation space to a lower correlation space. And the effects of mathematics changing and the geometric changing can be the same. Then the result can be gotten within a short observation period. The LLL algorithm decomposes a positive-definite symmetrical matrix into the upper/lower triangular matrix. Then uses the Gram–Schmidt orthogonalization to transform vectors of the matrix into orthogonal each other. Then the diagonal covariance matrix can be gotten by the transpose of the orthogonal matrix multiplying to the orthogonal matrix. Whitening filter uses crout factorization to decompose a positive-definite symmetrical matrix into the continue multiplication of diagonal matrix and unit upper/lower triangular matrix. Applying the specifics of its diagonal matrix condition to covariance matrix can get the diagonal covariance matrix. Using the diagonal covariance matrix can reduce the number of candidates for integral ambiguity. Final, the candidates are inserted into the observation equations to determine the solution again. It is believed that the integer candidate which produces the smallest sum of squares of the residual is the most likely solution we want.
