more comments, binary.

Hydra_EVSE/Hydra_EVSE.h

@@ -250,6 +250,13 @@
 // doe RB = 27 - Mega Hydra production
 //#define CURRENT_SCALE_FACTOR 184
 
+// Irregular-sampled EWA half-period for ammeter displays (the time after which data points a weighed 1/2
+// w.r.t. most recent data point). This is for Ammeter display only, it does not affect current measurements
+// for overdraw etc. purposes.
+#define AMM_DISPLAY_HALF_PERIOD 1500
+
+
+
 #define LOG_NONE 0
 #define LOG_INFO 1
 #define LOG_DEBUG 2
@@ -330,9 +337,6 @@ extern char p_buffer[];
 #include "units.h"
 #endif
 
-// EWA half-period for ammeter displays
-#define AMM_DISPLAY_HALF_PERIOD 3000
-
 struct car_struct {
   // CAR_A or CAR_B
   unsigned char car;

Hydra_EVSE/Hydra_EVSE.ino.standard.hex

@@ -1263,7 +1263,7 @@
 :104EE00010921503109216031092170310921803D4
 :104EF0001092190310921A0310921B0310921C03B4
 :104F000010921D0310921E0310921F031092210392
-:104F1000109220038EEA90E4A7E8B5E48093220380
+:104F1000109220038EEA90E4A7E0B5E48093220388
 :104F200090932303A0932403B093250310922E03A0
 :104F300010922F03109230031092310310922A0323
 :104F400010922B0310922C0310922D031092260323

Hydra_EVSE/onlineSum.h

@@ -21,8 +21,11 @@
 #ifndef ___ONLINESUM_H___
 #define ___ONLINESUM_H___
 
-// Exponentially Weighted Average summarizer
+///////////////////////////////////////////////////////////////////
+// Exponentially Weighted Average summarizer for irregularly sampled data.
+// References: 
 // http://tdunning.blogspot.com/2011/03/exponentially-weighted-averaging-for.html
+// My mods for the updates-in-the past (ensures no NaNs created regardless of updates order):
 // http://weatheringthrutechdays.blogspot.com/2011/04/follow-up-for-mean-summarizer-post.html
 
 // T is double or float
@@ -33,8 +36,14 @@ class EWASum {
     T w, s, tn;
 
   public:
-    // tHalfPeriod is amount of t after which the observation is exponentially weighed at 0.5 of the current observation.
-    EWASum(T tHalfPeriod) : alpha(-tHalfPeriod / log((T)0.5)) {
+
+    // tHalfWeightPeriod is the amount of t (timeline point) in the past at which the observation is 
+    // exponentially weighed at exactly 0.5 vs. any observation right now. Since it is weighed 
+    // exponentially rather than linearly, observations at full period still do have a nonzero weight, 
+    // and in general with this summarizer most recent observations are being discounted at fastest 
+    // weight but never actually discounted completely (indefinite history), down to the precision of
+    // the arithmetics. At full period the weights are thus still coming in at exp(-2) = 0.135 (13.5%).
+    EWASum(T tHalfWeightPeriod) : alpha(-tHalfWeightPeriod / log((T)0.5)) {
       reset();
     };
 
@@ -63,6 +72,9 @@ void EWASum<T>::update(T x, T t) {
 
 template <class T>
 T EWASum<T>::ewa() {
+  // In this sum, weights low enough should not happen, as we never discount the most recent weight, 
+  // which is at least one. So w>= 1 if there were at least one sample; and w = exactly 0 if there 
+  // are no samples at all. With no samples, we just return 0. 
   return (abs(w) < 1e-6) ? 0 : s / w;
 }