96c63ac3ec
had to change the parameter boundaries of the wifi optimizer to be able to use it for bluetooth... this should be refactored to something more generic.. some minor changes in ble
394 lines
13 KiB
C++
394 lines
13 KiB
C++
/*
|
||
* © Copyright 2014 – Urheberrechtshinweis
|
||
* Alle Rechte vorbehalten / All Rights Reserved
|
||
*
|
||
* Programmcode ist urheberrechtlich geschuetzt.
|
||
* Das Urheberrecht liegt, soweit nicht ausdruecklich anders gekennzeichnet, bei Frank Ebner.
|
||
* Keine Verwendung ohne explizite Genehmigung.
|
||
* (vgl. § 106 ff UrhG / § 97 UrhG)
|
||
*/
|
||
|
||
#ifndef WIFI_OPTIMIZER_LOG_DIST_CEILING_H
|
||
#define WIFI_OPTIMIZER_LOG_DIST_CEILING_H
|
||
|
||
#include "../../../floorplan/v2/Floorplan.h"
|
||
#include "../../../floorplan/v2/FloorplanHelper.h"
|
||
|
||
#include "../../../geo/BBox3.h"
|
||
#include "../../../misc/Debug.h"
|
||
|
||
#include "WiFiFingerprint.h"
|
||
#include "WiFiFingerprints.h"
|
||
|
||
#include "../model/WiFiModels.h"
|
||
//#include "../model/WiFiModelLogDistCeiling.h"
|
||
|
||
#include <KLib/math/optimization/NumOptAlgoDownhillSimplex.h>
|
||
#include <KLib/math/optimization/NumOptAlgoGenetic.h>
|
||
#include <KLib/math/optimization/NumOptAlgoRangeRandom.h>
|
||
|
||
#include <string>
|
||
#include <sstream>
|
||
|
||
#include "WiFiOptimizer.h"
|
||
#include "WiFiOptimizerStructs.h"
|
||
|
||
#include <functional>
|
||
|
||
namespace WiFiOptimizer {
|
||
|
||
/**
|
||
* optimize access-point parameters,
|
||
* given several fingerprints using the log-dist-ceiling model
|
||
*/
|
||
struct LogDistCeiling : public Base {
|
||
|
||
|
||
public:
|
||
|
||
/**
|
||
* resulting optimization stats for one AP
|
||
*/
|
||
struct Stats {
|
||
|
||
/** average model<->scan error after optimzing */
|
||
float error_db;
|
||
|
||
/** number of fingerprints [= locations] that were used for optimzing */
|
||
int usedFingerprins;
|
||
|
||
/** resulting model<->scan error after optimzing for each individual fingerprints [= location] */
|
||
std::vector<ErrorAtPosition> errors;
|
||
|
||
/** get the location where the model estimation reaches the highest negative value [model estimation too low] */
|
||
ErrorAtPosition getEstErrorMaxNeg() const {
|
||
auto cmpErrAtPos = [] (const ErrorAtPosition& a, const ErrorAtPosition& b) {return a.getError_db() < b.getError_db();};
|
||
return *std::min_element(errors.begin(), errors.end(), cmpErrAtPos);
|
||
}
|
||
|
||
/** get the location where the model estimation reaches the highest positive value [model estimation too high] */
|
||
ErrorAtPosition getEstErrorMaxPos() const {
|
||
auto cmpErrAtPos = [] (const ErrorAtPosition& a, const ErrorAtPosition& b) {return a.getError_db() < b.getError_db();};
|
||
return *std::max_element(errors.begin(), errors.end(), cmpErrAtPos);
|
||
}
|
||
|
||
};
|
||
|
||
using APFilter = std::function<bool(const Stats& stats, const MACAddress& mac)>;
|
||
|
||
static inline bool NONE(const Stats& stats, const MACAddress& mac) {
|
||
(void) stats; (void) mac;
|
||
return false;
|
||
}
|
||
|
||
static inline bool MIN_2_FPS(const Stats& stats, const MACAddress& mac) {
|
||
(void) mac;
|
||
return stats.usedFingerprins < 2;
|
||
}
|
||
|
||
static inline bool MIN_5_FPS(const Stats& stats, const MACAddress& mac) {
|
||
(void) mac;
|
||
return stats.usedFingerprins < 5;
|
||
}
|
||
|
||
static inline bool MIN_10_FPS(const Stats& stats, const MACAddress& mac) {
|
||
(void) mac;
|
||
return stats.usedFingerprins < 10;
|
||
}
|
||
|
||
|
||
|
||
/** parameters for one AP when using the LogDistCeiling model */
|
||
struct APParams {
|
||
|
||
float x;
|
||
float y;
|
||
float z;
|
||
|
||
float txp;
|
||
float exp;
|
||
float waf;
|
||
|
||
/** ctor */
|
||
APParams() {;}
|
||
|
||
/** ctor */
|
||
APParams(float x, float y, float z, float txp, float exp, float waf) : x(x), y(y), z(z), txp(txp), exp(exp), waf(waf) {;}
|
||
|
||
Point3 getPos() const {return Point3(x,y,z);}
|
||
|
||
std::string asString() const {
|
||
std::stringstream ss;
|
||
ss << "Pos:" << getPos().asString() << " TXP:" << txp << " EXP:" << exp << " WAF:" << waf;
|
||
return ss.str();
|
||
}
|
||
|
||
/** we add some constraints to the parameter range */
|
||
bool outOfRange() const {
|
||
return (waf > 0) ||
|
||
(txp < -65) ||
|
||
(txp > -45) ||
|
||
(exp > 5) ||
|
||
(exp < 1);
|
||
}
|
||
|
||
};
|
||
|
||
/** add MAC-info to params */
|
||
struct APParamsMAC {
|
||
MACAddress mac;
|
||
APParams params;
|
||
APParamsMAC(const MACAddress mac, const APParams& params) : mac(mac), params(params) {;}
|
||
};
|
||
|
||
struct OptResultStats {
|
||
K::Statistics<float> avgApErrors; // contains one average-error per optimized AP
|
||
K::Statistics<float> singleFPErrors; // contains one error for each fingerprint<->ap SIGNED
|
||
K::Statistics<float> singleFPErrorsAbs; // contains one error for each fingerprint<->ap ABSOLUTE
|
||
|
||
};
|
||
|
||
class APParamsList {
|
||
|
||
std::vector<APParamsMAC> lst;
|
||
|
||
public:
|
||
|
||
/** ctor */
|
||
APParamsList(const std::vector<APParamsMAC>& lst) : lst(lst) {
|
||
|
||
}
|
||
|
||
/** get the list */
|
||
const std::vector<APParamsMAC>& get() const {
|
||
return lst;
|
||
}
|
||
|
||
/** get params for the given mac [if known, otherwise nullptr] */
|
||
const APParamsMAC* get (const MACAddress& mac) const {
|
||
for (const APParamsMAC& ap : lst) {
|
||
if (ap.mac == mac) {return ≈}
|
||
}
|
||
return nullptr;
|
||
}
|
||
|
||
};
|
||
|
||
private:
|
||
|
||
Floorplan::IndoorMap* map;
|
||
|
||
Mode mode = Mode::QUALITY;
|
||
|
||
const char* name = "WiFiOptLDC";
|
||
|
||
public:
|
||
|
||
/** ctor */
|
||
LogDistCeiling(Floorplan::IndoorMap* map, const VAPGrouper& vg, const Mode mode = Mode::MEDIUM) : Base(vg), map(map), mode(mode) {
|
||
;
|
||
}
|
||
|
||
/** ctor */
|
||
LogDistCeiling(Floorplan::IndoorMap* map, const VAPGrouper& vg, const WiFiFingerprints& fps, const Mode mode = Mode::MEDIUM) : Base(vg), map(map), mode(mode) {
|
||
addFingerprints(fps);
|
||
}
|
||
|
||
/** optimize all known APs */
|
||
APParamsList optimizeAll(APFilter filter, OptResultStats* dst = nullptr) const {
|
||
|
||
// sanity check
|
||
Assert::isFalse(getAllMACs().empty(), "no APs found for optimization! call addFingerprint() first!");
|
||
|
||
K::Statistics<float> avgErrors;
|
||
K::Statistics<float> singleErrors;
|
||
K::Statistics<float> singleErrorsAbs;
|
||
|
||
std::vector<APParamsMAC> res;
|
||
for (const MACAddress& mac : getAllMACs()) {
|
||
|
||
// perform optimization, get resulting parameters and optimization stats
|
||
Stats stats;
|
||
const APParams params = optimize(mac, stats);
|
||
|
||
// filter based on stats (option to ignore/filter some access-points)
|
||
if (!filter(stats, mac)) {
|
||
res.push_back(APParamsMAC(mac, params));
|
||
//errSum += stats.error_db;
|
||
//++errCnt;
|
||
avgErrors.add(stats.error_db);
|
||
for (const auto e : stats.errors) {
|
||
singleErrors.add(e.getError_db());
|
||
singleErrorsAbs.add(std::abs(e.getError_db()));
|
||
}
|
||
} else {
|
||
Log::add(name, "ignoring opt-result for AP " + mac.asString() + " due to filter");
|
||
//std::cout << "ignored due to filter!" << std::endl;
|
||
}
|
||
|
||
}
|
||
|
||
//const float avgErr = errSum / errCnt;
|
||
//Log::add(name, "optimized APs: " + std::to_string(errCnt));
|
||
//Log::add(name, "average AP error is: " + std::to_string(avgErr) + " dB");
|
||
Log::add(name, "optimization result: ");
|
||
Log::add(name, " - AvgPerAP " + avgErrors.asString());
|
||
Log::add(name, " - Single: " + singleErrors.asString());
|
||
Log::add(name, " - SingleAbs: " + singleErrorsAbs.asString());
|
||
|
||
if (dst) {
|
||
dst->avgApErrors = avgErrors;
|
||
dst->singleFPErrors = singleErrors;
|
||
dst->singleFPErrorsAbs = singleErrorsAbs;
|
||
}
|
||
|
||
// done
|
||
return APParamsList(res);
|
||
|
||
}
|
||
|
||
|
||
/** optimize the given AP */
|
||
APParams optimize(const MACAddress& mac, Stats& res) const {
|
||
|
||
// starting parameters do not matter for the current optimizer!
|
||
APParams params(0,0,0, -59, 3, -8.0);
|
||
|
||
// get all position->rssi measurements for this AP to compare them with the corresponding model estimations
|
||
const std::vector<RSSIatPosition>& entries = apMap.find(mac)->second;
|
||
|
||
// log
|
||
Log::add(name, "optimizing parameters for AP " + mac.asString() + " by using " + std::to_string(entries.size()) + " fingerprints", false);
|
||
Log::tick();
|
||
|
||
// get the map's size
|
||
const BBox3 mapBBox = FloorplanHelper::getBBox(map);
|
||
|
||
using LeOpt = K::NumOptAlgoRangeRandom<float>;
|
||
const std::vector<LeOpt::MinMax> valRegion = {
|
||
LeOpt::MinMax(mapBBox.getMin().x - 20, mapBBox.getMax().x + 20), // x
|
||
LeOpt::MinMax(mapBBox.getMin().y - 20, mapBBox.getMax().y + 20), // y
|
||
LeOpt::MinMax(mapBBox.getMin().z - 6, mapBBox.getMax().z + 6), // z
|
||
LeOpt::MinMax(-65, -45), // txp
|
||
LeOpt::MinMax(1, 5), // exp
|
||
LeOpt::MinMax(-15, -0), // waf
|
||
};
|
||
|
||
|
||
|
||
|
||
LeOpt opt(valRegion);
|
||
|
||
switch(mode) {
|
||
case Mode::FAST:
|
||
opt.setPopulationSize(100);
|
||
opt.setNumIerations(50);
|
||
break;
|
||
case Mode::MEDIUM:
|
||
opt.setPopulationSize(200);
|
||
opt.setNumIerations(100);
|
||
break;
|
||
case Mode::QUALITY:
|
||
opt.setPopulationSize(1500);
|
||
opt.setNumIerations(150);
|
||
break;
|
||
}
|
||
|
||
// error function
|
||
auto func = [&] (const float* params) {
|
||
return getErrorLogDistCeiling(mac, entries, params, nullptr);
|
||
};
|
||
|
||
opt.calculateOptimum(func, (float*) ¶ms);
|
||
|
||
// using LeOpt = K::NumOptAlgoGenetic<float>;
|
||
// LeOpt opt(6);
|
||
// opt.setPopulationSize(750);
|
||
// opt.setMaxIterations(50);
|
||
// opt.setElitism(0.05f);
|
||
// opt.setMutation(0.75f);
|
||
// //opt.setValRange({0.5, 0.5, 0.5, 0.1, 0.1, 0.1});
|
||
// opt.setValRegion(valRegion);
|
||
|
||
// K::NumOptAlgoDownhillSimplex<float, 6> opt;
|
||
// opt.setMaxIterations(100);
|
||
// opt.setNumRestarts(10);
|
||
|
||
opt.calculateOptimum(func, (float*) ¶ms);
|
||
res.error_db = getErrorLogDistCeiling(mac, entries, (float*)¶ms, &res);
|
||
res.usedFingerprins = entries.size();
|
||
|
||
Log::tock();
|
||
Log::add(name, mac.asString() + ": " + params.asString() + " @ " + std::to_string(res.error_db) +" dB err");
|
||
|
||
return params;
|
||
|
||
}
|
||
|
||
|
||
private:
|
||
|
||
float getErrorLogDistCeiling(const MACAddress& mac, const std::vector<RSSIatPosition>& entries, const float* data, Stats* stats = nullptr) const {
|
||
|
||
const APParams* params = (APParams*) data;
|
||
|
||
// some sanity checks
|
||
if (params->outOfRange()) {return 1e10;}
|
||
|
||
// current position guess for the AP;
|
||
const Point3 apPos_m = params->getPos();
|
||
|
||
// add the AP [described by the current guess] to the signal-strength-prediction model
|
||
// signal-strength-prediction-model...
|
||
WiFiModelLogDistCeiling model(map);
|
||
model.addAP(mac, WiFiModelLogDistCeiling::APEntry(apPos_m, params->txp, params->exp, params->waf));
|
||
|
||
float err = 0;
|
||
int cnt = 0;
|
||
|
||
// process each measurement
|
||
for (const RSSIatPosition& reading : entries) {
|
||
|
||
// get the model-estimation for the fingerprint's position
|
||
const float rssiModel = model.getRSSI(mac, reading.pos_m);
|
||
|
||
// difference between estimation and measurement
|
||
const float diff = std::abs(rssiModel - reading.rssi);
|
||
|
||
// add error to stats object?
|
||
if (stats) {
|
||
stats->errors.push_back(ErrorAtPosition(reading.pos_m, reading.rssi, rssiModel));
|
||
}
|
||
|
||
// adjust the error
|
||
err += std::pow(std::abs(diff), 2.0);
|
||
++cnt;
|
||
|
||
// max distance penality
|
||
// [unlikely to get a reading for this AP here!]
|
||
if (apPos_m.getDistance(reading.pos_m) > 150) {err += 999999;}
|
||
|
||
}
|
||
|
||
err /= cnt;
|
||
err = std::sqrt(err);
|
||
|
||
if (params->txp < -65) {err += 999999;}
|
||
if (params->txp > -45) {err += 999999;}
|
||
|
||
if (params->exp > 5) {err += 999999;}
|
||
if (params->exp < 1.0) {err += 999999;}
|
||
|
||
return err;
|
||
|
||
}
|
||
|
||
|
||
|
||
};
|
||
|
||
}
|
||
|
||
|
||
#endif // WIFI_OPTIMIZER_LOG_DIST_CEILING_H
|