Skip to content
rtc_access_object.py 25.4 KiB
Newer Older
#!/usr/bin/env python
# -*- coding: utf-8 -*-

vertighel's avatar
vertighel committed
import json
import sys
import datetime

import numpy as np
from astropy.time import Time
vertighel's avatar
vertighel committed
from util import functions as ufun
from util import variables as uvars
from util.master_listener_davide import thread
from util.mini_os import obsController
vertighel's avatar
vertighel committed

### Focus only on the 8x8 pixel matrix centered on the centroid
def focus_on_centroid(a, cx, cy, s=4):
    # a is an array-like matrix , s is an Int
    cx = int(cx)*32 + 32   # cx and cy are normalized in [-1,1] -> now are renormalized in [0,64]
    cy = int(cy)*32 + 32

    if(s>cx or s>cy): 
        s = min(cx,cy)
        
    if(s>len(a)-cx or s>len(a)-cy):
        s = min(s , len(a)-cx , len(a)-cy)
        
    a_focused = a[cx-s : cx+s , cy-s : cy+s]
    return a_focused
class RtcObject(object):
    def __init__(self):
        self.counter = 0
        self.todos = []
        self.rotating_ncpa = False
        self.ncpa_file = "No file"
        self.json_state = {}

        self.thread = thread
        self.obsController = obsController
    def reset(self):
        print("reset")
        status = self.obsController.RTCTTResetBCU()
        return status
    def resurrect(self):
        print("import")
        from SHINS_SW_TEST.SHINS_TEST_DEBUG import SHINS_TEST_DEBUG
        
        print("init")        
        status = SHINS_TEST_DEBUG(test_s="10", test_s1="true") # init
        time.sleep(1)
        print("setup start")
        status = SHINS_TEST_DEBUG(test_s="11") # setup and start
        print(status)
        return status

    def resurrect2(self):
        print('resurrect2')
        rtc_s = uvars.rtc_s_base.copy()
        time.sleep(10)
        print("init")
        status = self.obsController.RTCTTInit(rtc_s)
        print("Exit status:  {}".format(status))
        ufun.error_handler(status, "RTCTTIsInit")
            
        print("Wait 10 s...")
        time.sleep(10)
        # SETUP REAL-TIME RTCTT #

        print("RTCTT setup... Wait 15 s...")
        status = self.obsController.RTCTTSetup()
        print("Exit status:  {}".format(status))
        ufun.error_handler(status, "RTCTTSetup")
        
        # START REAL-TIME RTCTT #
        print("RTCTT start")
        time.sleep(10)
        status = self.obsController.RTCTTStart()
        print("Exit status:  {}".format(status))
        ufun.error_handler(status, "RTCTTStart")

    def uploadRTC(self, modal=True):
        rtc_s = uvars.rtc_s_base.copy()
        if modal:
            print("Enabling Modal Mode")
        else:
            print("Enabling Zonal Mode")
            rtc_s["TTM2CFILE"] = "ZONAL"
        
        # INIT REAL-TIME RTCTT #
        print("Enumerating RTC setup")
        for element, value in rtc_s.items():
            print("Element : % 20s  -  Value % 20s    " % (element, value))
        
        status = self.obsController.RTCTTInit(rtc_s)
        print("Exit status:  {}".format(status))
        ufun.error_handler(status, "RTCTTIsInit")
            
        print("Wait 5 s...")
        time.sleep(5)
        # SETUP REAL-TIME RTCTT #

        print("RTCTT setup... Wait 15 s...")
        status = self.obsController.RTCTTSetup()
        print("Exit status:  {}".format(status))
        ufun.error_handler(status, "RTCTTSetup")
        
        # START REAL-TIME RTCTT #
        print("RTCTT start")
        status = self.obsController.RTCTTStart()
        print("Exit status:  {}".format(status))
        ufun.error_handler(status, "RTCTTStart")

        mode = "Modal" if modal else "Zonal"
        print("Enabling {} Mode... DONE".format(mode))
        
    def zonalCommand(self, data):
        print("chiamata effettuata")
        normalize_strain = 1#6.666667
        act = int(data[0])
        strain = float(data[1])
        print(strain)
        strain = strain/normalize_strain
        print("Act on actuator number {} with absolute intensity {}".format(act,strain))
        reserved = [0,0]
        
        actuators = np.zeros(97).tolist()
        actuators[act-1] = strain
        
        print("vector for RTCTTModeUpload: ")
        # print(actuators)
        try:
            status = self.obsController.RTCTTModeUpload(reserved+actuators, absolute=False)
        except Exception as e:
            print(e)
            status = 1
        return status
    def move_soul(self, data, apply_movement=False):
        self.soul_deltas = data
        print("Received data: {}".format(data))
        PSFcx  = float(data[0])
        PSFcx  = 0    if PSFcx < 0    else PSFcx
        PSFcx  = 2048 if PSFcx > 2048 else PSFcx 
        PSFcy  = float(data[1])
        PSFcy  = 0    if PSFcy < 0    else PSFcy
        PSFcy  = 2048 if PSFcy > 2048 else PSFcy
        
        deltaX = PSFcx - uvars.SOUL_TARGET_POS[0]
        deltaY = PSFcy - uvars.SOUL_TARGET_POS[1]
        print("Calling Here OffsetXYAO with SCICAM deltas Dx = {} , Dy = {}".format(deltaX , deltaY))

        # GET CURRENT DROT #
        status, drotDict = self.obsController.getPosition("DEGREE", {"DROT":-9999.9})
        drotPos = drotDict["DROT"]
        print("DROT pos is {}".format(drotPos))
        
        # Transform to SOUL delta #
        step_x, step_y = ufun.transform_offset(disp = [deltaX,deltaY], thetaRot = drotPos , conv = "SOUL")
        print("Calculatd steps to apply to SOUL : {} , {}".format(step_x , step_y))
        if apply_movement:
            max_step = 1
            step_x = step_x if (abs(step_x) <= max_step) else max_step*np.sign(step_x)
            step_y = step_y if (abs(step_y) <= max_step) else max_step*np.sign(step_y)
            print("Applying these steps to SOUL : {} , {}".format(step_x , step_y))
            try:
                status = self.obsController.OffsetXYAO(step_x , step_y , "LEFT")
                print("Function is returning NOW")

            except Exception as e:
                print(e)
                status = 1

        else:
            from util.shinsLogger import logger
            resume = "Delta X = {} mm  ,  Delta Y = {} mm".format(step_x,step_y)
            logger.info("--------------------------------------------------------------------------")
            logger.info("| {:^70s} |".format(resume))
            logger.info("--------------------------------------------------------------------------")
            logger.debug("SCICAM deltas = ({},{}), DROT pos = {}, computed SOUL deltas = ({},{})".format(deltaX,deltaY,drotPos,np.round(step_x,3),np.round(step_y,3)))
        return status, [np.round(step_x,4), np.round(step_y,4)] 


    def manageDerotation(self, data):
        action  = str(data[0])
        try:
            mode    = str(data[1])
            pos_ang = float(data[2])
            pos_ang = pos_ang-360 if pos_ang>180 else pos_ang
        except Exception as e:
            # print(e)
            mode, pos_ang = None, None
        
        if(action == "Start_Tracking"):
              print("DAO_______________Derotation mode: {} with angle {}".format(mode,pos_ang))
              ins_s = {"DROT_Mode":mode}
              async = self.obsController.begin_setupInstrument(ins_s)
              status = self.obsController.end_setupInstrument(async)
                      
              if mode == "FIELD_FREE":
                  print("Launching now a {} derotation".format(mode))
                  status = self.obsController.trackingStart2(0,0)
                  
              elif mode == "FIELD_FIXED":
                  print("Launching now a {} derotation".format(mode))
                  dateMjd= Time.now()
                  status = self.obsController.trackingStart2(dateMjd.mjd, pos_ang)
                  
              else:
                  print("{} mode: NO derotation".format(mode))
                  status = self.obsController.trackingStop()
                  
        elif(action == "Stop_Tracking"):
              print("DAO__________________________STOP Derotation!")
              status = self.obsController.trackingStop()
        else:
              print("Something went wrong, no action received")
        print("Reached end of command")
        return status
    def read(self, id):
        for todo in self.todos:
            if todo["id"] == id:
                return todo
            
    def create(self, data):
        todo = data
        self.todos.append(todo)
        return todo

    def update(self, id, data):
        todo = self.read(id)
        todo.update(data)
        return todo

    def delete(self, id):
        todo = self.read(id)
        self.todos.remove(todo)

    @property
    def centroid(self):
        status, [cx, cy] = self.obsController.getInternalSeqFloatVariable("RtcCentroidOrigin")
        return [cx, cy]

    @centroid.setter
    def centroid(self, cx_cy):
        cx, cy = cx_cy
        status = self.obsController.RTCTTSetCentroidPos(cx, cy)
        return status

    def flatten(self):
        modes = [0] * 96#39
        status = self.obsController.RTCTTModeUpload(modes, absolute=True)
        status = self.load_ncpa(name="zeri.txt")
        return status
    def close_loop(self):
        status = self.obsController.RTCTTCloseLoop()
        return status

    def open_loop(self):
        status = self.obsController.RTCTTOpenLoop()
        return status

    def frequency(self, freq):
        """Set the loop frequency"""

        rtc_s = {"TTFRAMERATE": freq}
        status = self.obsController.RTCTTExecuteSetup(rtc_s)

        return status
    def tint(self, t):
        """Set integration time"""

        rtc_s = {
            "TTCAMTINT" : t,
        }
        status = self.obsController.RTCTTExecuteSetup(rtc_s)
        return status

    def patrol(self, b):
        """Set params to enable or disable patrol camera mode"""

        if b:
            print("Switch to patrol mode")
            rtc_patrol = {
                "TTM2CFILE": "20220603-Zernike-FullPupil-Arcetri-1u-rms-4rtc.dat",
                "TTWINCOORDX": "0",
                "TTWINCOORDY": "0",
                "TTWINCOLS": "640",
                "TTWINROWS": "512",
                "TTNSLOPEPIXELS": "16",
                "TTBIASFILE": "",
                "TTPIXELGAINFILE": "",
                "TTCAMTINT": "0.004995574",
                "TTPIXELGAINMODE": "FLATGAIN",
            }
            
        else:
            print("Switch to cropped mode")
            rtc_small = {
                "TTM2CFILE": "20220603-Zernike-FullPupil-Arcetri-1u-rms-4rtc.dat",
                "TTWINCOORDX": "224",
                "TTWINCOORDY": "212",
                "TTWINCOLS": "64",
                "TTWINROWS": "64",
                "TTFRAMERATE": "1000",
                "TTNSLOPEPIXELS": "4096",
                "TTBIASFILE": "2022-02-19-23-45-23_PixelBias_19.95°C_1000Hz-064x064.dat",
                "TTPIXELGAINFILE": "flat_gain_64x64.dat",
                "TTCAMTINT": "0.000994919",
                "TTPIXELGAINMODE": "FLATGAIN",
            }

        rtc_s = rtc_patrol if b else rtc_small
        
        status = self.obsController.RTCTTSwitchCcdMode(rtc_s)

        return status


    def temperature(self, temp):
        """Set the loop temperature"""

        status = self.obsController.RTCTTSendCameraCommand("set temperatures sensor "+temp)

        return status


    def pid(self, p):
        """Set the RTC PID"""
        
        rtc_s = {
            "TTPIDKP": str(p[0]),
            "TTPIDKI": str(p[1]),
            "TTPIDKD": str(p[2]),
            "TTPIDTF": str(p[3]),
            "TTPIDTS": str(p[4]),
            }
        rtc_s["TTPIDTS"]=str(1e-3)
        rtc_s["TTPIDTF"]=rtc_s["TTPIDTS"]
        print(rtc_s)
        status = self.obsController.RTCTTExecuteSetup(rtc_s)
        
        return status

    def threshold(self, t):
        """Set the RTC Threshold"""
        
        status = self.obsController.RTCTTWriteSingle(4416, t, "dsp")
        
        return status

    def get_threshold(self):
        """Get the RTC Threshold"""
        
        status, threshold = self.obsController.RTCTTReadSingle(4416, "dsp")
        
        return threshold

        
    ##############
    # Dark stuff #
    ##############

    def todos_dark(self):
        """List dark files"""

        return [ str(dark.name).decode('utf-8') for dark in sorted(uvars.dark_path.glob("*.dat")) ]

    def new_dark(self):
        """Create a new dark"""

        rtc_s = {
            "TTDIAGENABLED" : "true", # or "1"
            "TTDIAGDECIMATION" : "0",
            "TTPIXELENABLED" : "1",
            "TTPIXELDECIMATION" : "0", # if 0 master diagnostic stops
            "TTLOOPENABLED" : "false",
            "TTSAVEASBIAS" : "true",
            "TTUPDATEBIAS" : "true",
            "TTTESTTIME" : "1000000",
        }
        status, filename = self.obsController.RTCTTTeccamExpose(rtc_s)

        return filename

    def load_dark(self, name):
        """Load a dark file"""
        rtc_s = {"TTBIASFILE": name}
        status = self.obsController.RTCTTExecuteSetup(rtc_s)

        return status

    def get_current_dark(self):
        pass

    ##############
    # Gain stuff #
    ##############

    def todos_gain(self):
        """List gain files"""

        return [ str(gain.name).decode('utf-8') for gain in sorted(uvars.gain_path.glob("*.dat")) ]

    def new_gain(self, gain_type, centroid_size):
        """Create a new gain"""
        rtc_s = {
            "TTPIXELGAINMODE": gain_type,
            "TTPIXELGAINRADIUS": centroid_size,
        }
        status = self.obsController.RTCTTExecuteSetup(rtc_s)
        return status

    def load_gain(self, name):
        """Load a gain file"""
        rtc_s = {
            "TTPIXELGAINMODE": "FILEGAIN",
            "TTPIXELGAINFILE": name,
        }
        status = self.obsController.RTCTTExecuteSetup(rtc_s)
        return status

    ##############
    # Flat stuff #
    ##############

    def todos_flat(self):
        """List flat files"""
        return [ str(flat.name).decode('utf-8') for flat in sorted(uvars.flat_path.glob("*.txt")) ]
    
    
    def load_flat(self, name):
        """Load a flat file"""
        rtc_s = {"TTDMFLATFILE": name}
        status = self.obsController.RTCTTExecuteSetup(rtc_s)
        return status

    def tiptilt(self, modes=[0,0,0,0], absolute=False):
        print("Loaded Modes {}".format(modes))
        status = self.obsController.RTCTTModeUpload(modes=modes, absolute=absolute)
        return status

    def teccam_custom_command(self, command):
        status, res = self.obsController.RTCTTSendCameraCommand(command)
        return res
    
    def save(self):
        data = self.data()
        now = Time.now().isot

        filename = "/data/rtc/rtc-panel-save-{}.json".format(now)
        with open(filename, 'w') as f:
            json.dump(data, f)
        return filename
    
    def save_flat(self):
        """Save the current flat in a file"""

        print("hereeee")
        flat_name = ufun.save_flat(self.obsController, self.thread.shape)
        print("doneee")
        print(flat_name)
        return flat_name

    def cloud_samples(self, s):
        print("prima : samples = {}".format(self.thread.cloud_samples))
        status = self.thread.cloud_samples = s
        print("dopo : samples = {}".format(self.thread.cloud_samples))
        return status

    def get_cloud_samples(self):
        return self.thread.cloud_samples

    ##############
    # NCPA stuff #
    ##############

    def todos_ncpa(self):
        """List ncpa files"""
        return [ str(ncpa.name).decode('utf-8') for ncpa in sorted(uvars.ncpa_path.glob("*.txt")) ]

    def load_ncpa(self, name):
        """Load a ncpa file"""
        # load saved file

        mode_file = open(str(uvars.ncpa_path)+name, 'r')
        data = mode_file.read()
        data_into_list = data.split('\n')
        mode_vector = [float(element) for element in data_into_list[:-1]]

        # edit mode vector obtained from loaded file
        print("v=[{}], len = {}".format(mode_vector, len(mode_vector)))
        angle_where_ncpa_was_optimized = float(mode_vector.pop())
        print("angle read from ncpa file: {}".format(angle_where_ncpa_was_optimized))
        print("v=[{}], len = {}".format(mode_vector, len(mode_vector)))
        mode_vector = mode_vector[2:]
        print("final mode vector=[{}], len = {}".format(mode_vector, len(mode_vector)))

        #apply correct rotation (if necessary)
        actual_drot_pos = float(self.obsController.getPosition("DEGREE", {"DROT":-9999})[1]["DROT"])
        drot_diff = actual_drot_pos -angle_where_ncpa_was_optimized

        if(abs(drot_diff)>0.5):
            ruotati = ufun.rotate_modes(mode_vector, -1*drot_diff)
        else:
            ruotati = mode_vector
        riservati = [0,0]
        try:
            status = self.obsController.RTCTTModeUpload( riservati + list(ruotati), absolute=True)
        except Exception as e:
            print(e)
            status = 1
        self.obsController.setContext({"NCPA_FILE":name});
        print("Finished")
        return status

    def save_ncpa(self):
        """Save the current ncpa in a file"""
        ncpa_name = ufun.save_ncpa(os=self.obsController)
        print(ncpa_name)
        return ncpa_name

    
    def getNCPARotationStatus(self):
        print("Status ncpa rotation in arrivo... status is {}".format(self.rotating_ncpa))
        return self.rotating_ncpa

    
    def rotateNCPA(self, action):
        
        if action == "start":
            print("START NCPA Rotation")
            self.rotating_ncpa = True
            self.obsController.setContext({"NCPA_ROT_STATUS":"Rotation ON"});
        elif action == "stop":
            print("STOP NCPA Rotation")
            self.rotating_ncpa = False
            self.obsController.setContext({"NCPA_ROT_STATUS":"Rotation OFF"});
        else:
            print("Wrong command in rotateNCPA")

        riservati = [0,0]
        sign = -1 # rotation direction
        sleep = 5#10
        threshold = 0.5

        ins_s = {"DROT":-9999}
        angle_where_ncpa_was_optimized = float(self.obsController.getPosition("DEGREE", ins_s)[1]["DROT"])

        status, total_vector = self.obsController.getInternalSeqFloatVariable("RtcAppliedMode")

        mode_vector = total_vector[2:]
        position_of_last_update = angle_where_ncpa_was_optimized
        
        while self.rotating_ncpa:
            
            bearing = float(self.obsController.getPosition("DEGREE", ins_s)[1]["DROT"])
            diff = bearing - position_of_last_update
            
            print("Bearing: {}; Diff wrt previous update: {}".format(bearing, diff))
            if abs(diff) > threshold:
                print("Above threshold: rotating modes!")
                ruotati = ufun.rotate_modes(mode_vector, sign * (bearing - angle_where_ncpa_was_optimized) )[0:36]
                self.obsController.RTCTTModeUpload( riservati + list(ruotati), absolute=True)
                position_of_last_update = bearing
            else:
                print("Below threshold: no mode rotation")            
                
            time.sleep(sleep)
        return 1



    @property
    def rtc_image(self):
        
        dummy_64 = np.zeros(shape=(64,64))

        # Diagostic stuff
        try:
            #print("trying to get the image")
            image = self.thread.img.T # transpose
            #print("there is an image. probably diagnostic is arriving")
        except AttributeError as e:
            print("no image")
            print("if NO image but diagnostic data from os are coming then try reset and resurrect rtc")
            print("can also try to request fps to the console on rtc panel, and then again reset+resurrect")
            image = dummy_64
        except IndexError as e:
            print("too many indices")
            image = dummy_64
        except Exception as e:
            print("Generic error")
            print(e)

        return image


    def data_cred(self, image, centroid):
        # # OS

        status, darkarray = self.obsController.getInternalSeqFloatVariable("RtcPixelBias")

        try:            
            darkimage = np.array(darkarray).reshape(image.shape).T
        except Exception as e:
            darkimage = np.zeros(shape=(image.shape))
        
        try:
            cx,cy = centroid
            image = image[2:-2 , 2:-2]
            darkimage = darkimage[2:-2 , 2:-2]
            darkedData  = image - darkimage
            cred = {
                "data": image.tolist(),
                "dark": darkimage.tolist(),
                #"darkedData" = darkedData,
                "min": image.min(),              #image[1:-1, 1:-1].min(),
                "max": image.max(),              #image[1:-1, 1:-1].max(),
                "minDark" : darkedData.min(),    #darkedData[3:, 3:].min(),
                "maxDark" : darkedData.max(),    #darkedData[3:, 3:].max(),
                "cx": cx,
                "cy": cy,
                "image_shape": list(image.shape),
            }

        except Exception as e:
            print("error building the cred response")
            print(e)
            print("image shape : {} ".format(image.shape))
            #print("image shape : {}  ,  darkedData shape : {}".format(image.shape, darkedData.shape))
            
            cred = None
            
        return cred


    def data_gain(self, imshape):

        # OS
        status, pixgain = self.obsController.getInternalSeqFloatVariable("RtcPixelGain")

        try:            
            pixgain = np.array(pixgain).reshape(imshape).T
        except Exception as e:
            pixgain = np.zeros(shape=imshape)+1
                "data": pixgain.reshape(imshape).tolist(),
                "min": pixgain[3:, 3:].min(),
                "max": pixgain[3:, 3:].max(),
                "image_shape": list(imshape),
            }

        except Exception as e:
            print("error building the gain response")
            print(e)
            gain = None

        return gain

    def data_dm(self):

        signs_list = []
        with open(str(uvars.sign_map), "r") as f:
            for line in f:
                for x in line.split():
                    signs_list.append(int(x))
        signs = np.array(signs_list)

        # Diagnostics
        shape = self.thread.shape

        # OS
        status, dmflat = self.obsController.getInternalSeqFloatVariable("RtcDmFlat")
        
        try:
            dm_shape_no_flat = (shape*signs) - (np.array(dmflat[0:97]))
            dm = {
                "actuators": uvars.actuators.tolist(),
                "shape": (shape*signs).tolist(),
                "flat": (np.array(dmflat[0:97])).tolist(),
                "min": round(shape.min(),2),
                "max": round(shape.max(),2),
                "min_no_flat" : round(min(dm_shape_no_flat),2),
                "max_no_flat" : round(max(dm_shape_no_flat),2),
                "skp_cmd"    : self.thread.tt_cmd[1]
            }

        except Exception as e:
            print("error building the dm response")
            print(e)
            dm = None

        return dm


    def data_cloud(self, imshape, centroid):
        centroids_x = self.thread.centroid_x
        centroids_y = self.thread.centroid_y
        cloud = np.flip(self.thread.centroid_matrix.T, 1)
        cx,cy = centroid
        rmsX = np.std(centroids_x)*21  # px 2 mas,  plate scale = 21 mas/px
        rmsY = np.std(centroids_y)*21  # px 2 mas
        wx = (np.max(centroids_x)-np.min(centroids_x))*21 # px to mas
        wy = (np.max(centroids_y)-np.min(centroids_y))*21 # px to mas
        
        try:

            cld = {
                "data": cloud.tolist(),
                "min": cloud.min(),
                "max": cloud.max(),
                "samples": len(self.thread.centroid_x),
                "cx": cx - self.thread.cx,
                "cy": cy - self.thread.cy,
                "wx": round(wx, 5), 
                "wy": round(wy, 5), 
                "image_shape": list(imshape),
                "RMS_X": round(rmsX, 5),
                "RMS_Y": round(rmsY, 5),
                "cloud_samples": self.thread.cloud_samples,
            }

        except Exception as e:
            print("error building the cloud response")
            print(e)
            cld = None

        return cld

        image = self.rtc_image
        imshape = image.shape
        centroid = self.centroid

        data_cloud = self.data_cloud(imshape, centroid)
        data_cred = self.data_cred(image, centroid)
        
        
        cropped_cred = focus_on_centroid( np.array(data_cred["data"]) , data_cloud["cx"] ,data_cloud["cy"], s=4)

        flux_cred = np.sum(cropped_cred)
                
        data_cred["tot_flux"] = flux_cred   # added data flux to jso

        cropped_cred_darked = focus_on_centroid( np.array(data_cred["data"])-np.array(data_cred["dark"]) , data_cloud["cx"] ,data_cloud["cy"], s=4)

        flux_cred_darked = np.sum(cropped_cred_darked)
        
        data_cred["tot_flux_darked"] = flux_cred_darked  # added darked flux to json
        
        response = {
            "cred": data_cred,
            "gain": self.data_gain(imshape),
            "dm": self.data_dm(), # removed imshape as parameter
            "cloud": data_cloud,
            "cx": centroid[0],
            "cy": centroid[1],
        }

        #print(response["cred"]["tot_flux"])

        self.json_state = response
        return response