chore: update README.md

speedtest/internal/README.md

@@ -1,11 +1,11 @@
-# Issue 192
+# Issue #192
 
-<img width="1074" alt="SpeedTest-Go" src="https://github.com/showwin/speedtest-go/assets/30739857/203da134-77f7-46a1-bed2-2df3c43ffc7c">
+<img width="1172" alt="SpeedTest-Go (1)" src="https://github.com/showwin/speedtest-go/assets/30739857/ecced0e9-830c-42d6-aa8b-e3dcf8d124d6">
 
 1. Use welford alg to quickly calculate standard deviation and mean.
-2. (a)The welford method integrated moving window feature, This allows us to ignore early data with excessive volatility.
-3. (b)The welford method integrated moving Average feature, compare with 2a, this causes early data to be diluted gradually rather than immediately.
-4. Use the coefficient of variation(c.v) to reflect the confidence of the test result datasets.
-5. When the c.v is less than 0.05, we terminate this test.
-
-Is there a better way?
+2. The welford alg integrated moving window feature, This allows us to ignore early data with excessive volatility.
+3. Use the coefficient of variation(c.v) to reflect the confidence of the test result datasets.
+4. When the data becomes stable(converge), the c.v value will become smaller. When the c.v < 0.05, we terminate this test. We set the tolerance condition as the window buffer being more than half filled and triggering more than five times with c.v < 0.05.
+5. Perform EWMA operation on real-time global average, and use c.v as part of the EWMA feedback parameter.
+6. The ewma value calculated is the result value of our test.
+7. When the test data converge quickly, we can stop early and speed up the testing process. Of course this depends on network/device conditions.