diff --git a/src/changes/changes.xml b/src/changes/changes.xml
index 0c7b38e1651859a2b1405b6dd98458be37b202d0..673e6c311c2869f0f7c29ec2c2f46227df0b5ba8 100644
--- a/src/changes/changes.xml
+++ b/src/changes/changes.xml
@@ -30,6 +30,9 @@
       <action dev="guylaine" type="update" issue="390">
         Changed CS Group website URL.
       </action>
+      <action dev="guylaine" type="update">
+        Updated link to Sergeï Tanygin's paper about attitude interpolation in documentation.
+      </action>
     </release>
     <release version="2.2" date="2020-07-31" description="This is a minor release.
      It adds access to algorithm identifier, 
diff --git a/src/site/markdown/design/technical-choices.md b/src/site/markdown/design/technical-choices.md
index 182f554604703e0467aada2484948b5cd449ea49..58233b3d53ce6bd2ceeb4553e6cf6346849274ae 100644
--- a/src/site/markdown/design/technical-choices.md
+++ b/src/site/markdown/design/technical-choices.md
@@ -105,7 +105,7 @@ components between Q1 and Q2 or -Q1 and Q2 leads to completely different rotatio
 will typically have one sign change per orbit at some random point. The third reason is that instead of doing an
 interpolation that respect quaternions constraint, the interpolation departs from the constraint first and attempts to
 recover afterwards in a normalization step. Orekit uses a method based on Sergeï Tanygin's paper
-[Attitude interpolation](http://www.agi.com/resources/white-papers/attitude-interpolation) with slight
+[Attitude interpolation](https://www.agi.com/resources/whitepapers/attitude-interpolation) with slight
 changes to use modified Rodrigues vectors as defined in Malcolm D Shuster's
 [A Survey of Attitude Representations](http://www.ladispe.polito.it/corsi/Meccatronica/02JHCOR/2011-12/Slides/Shuster_Pub_1993h_J_Repsurv_scan.pdf),
 despite attitude is still represented by quaternions in Orekit (Rodrigues vectors are used only for interpolation).