def closest_matrix_from_V(vpos_mat, spos_mat, c2pos_mat, c3pos_mat, inputfile):
close_array = np.array([1000])
v = vpos_mat[1:]
c2 = c2pos_mat[1:]
c3 = c3pos_mat[1:]
s = spos_mat[1:]
n = len(c2) / (len(s))
f = len(s) / len(v)
for i in range(0, (len(v))):
a = np.array([float(v[i][0]), float(v[i][1]), float(v[i][2])])
for su in range(0, f):
#b=np.array([float(s[su+i*f][0]),float(s[su+i*f][1]),float(s[su+i*f][2])])
#close_array=np.append(close_array,[u.length(np.subtract(a,b))],axis=0)
for x in range(0, n):
b = np.array([
float(c2[x + n * su + n * f * i][0]),
float(c2[x + n * su + n * f * i][1]),
float(c2[x + n * su + n * f * i][2])
])
close_array = np.append(close_array,
[u.length(np.subtract(a, b))],
axis=0)
b = np.array([
float(c3[su + i * f][0]),
float(c3[su + i * f][1]),
float(c3[su + i * f][2])
])
close_array = np.append(close_array, [u.length(np.subtract(a, b))],
axis=0)
return close_array
def angle_between(v, s, c):
normal = np.subtract(s, v)
inter = np.subtract(c, s)
theta = np.arccos(
np.dot(normal, inter) / (u.length(normal) * u.length(inter)))
theta = np.degrees(theta)
return theta
def angle_between(vec1, vec2):
x = np.dot(vec1, vec2) / (u.length(vec1) * u.length(vec2))
if (x > 1.0):
x = 1.0
#print u.length(vec1)
#print u.length(vec2)
theta = np.arccos(x)
theta = np.degrees(theta)
#print theta
return theta
def collide_guides(self, car):
# guide vectors:
vectors = car.guidelines_centered()
intersects = []
(outer, inner) = self.getlines()
outersegments = self._line_segments(outer)
innersegments = self._line_segments(inner)
p = []
for v in vectors:
vectorpoints = []
vectorpoints += self._collide_lines(v, outersegments)
vectorpoints += self._collide_lines(v, innersegments)
if len(vectorpoints) > 0:
minlength = 10000000
point = vectorpoints[0]
for x in vectorpoints:
l = util.length(car.rect.center, x)
if (l < minlength):
minlength = l
point = x
p += [point]
return p
def move(self, e):
pos = np.array([e.x, e.y],dtype=float)
centred = pos-self.size/2
new_colour = self.colour.copy()
if self.selected == 0:
self.canvas.delete(self.cursors[1])
pos[1] = max(10,min(pos[1],self.size-10))
self.move_cursor(1, self.size+10, pos[1])
val = 1 - (pos[1]-10) / (self.size-20)
self.refresh(val)
new_colour[2] = val
elif self.selected == 1:
sat = util.length(centred) / self.size * 2
if sat > 1:
self.move_cursor(0, *(centred*(1/sat) + self.size/2))
else:
self.move_cursor(0, *pos)
new_colour[0] = (math.atan2(centred[1], centred[0]) / math.pi / 2) % 1
new_colour[1] = min(sat, 1)
else:
return
self.update_textboxes(new_colour)
def position_matrix_from_V(vpos_mat,Aupos_mat, spos_mat,c2pos_mat,c3pos_mat, inputfile):
close_array=np.array([[100,100,100]])
v=vpos_mat[1:]
au=Aupos_mat[1:]
c2=c2pos_mat[1:]
c3=c3pos_mat[1:]
s=spos_mat[1:]
thetax=0
thetay=0
thetaz=0
correct=[0,0,2]
n=len(c2)/(len(s))
f=len(s)/len(v)
for i in range(0,(len(v))):
a=np.array([float(v[i][0]),float(v[i][1]),float(v[i][2])])
for d in range(i*122,(i+1)*122):
e=np.array([float(au[d][0]),float(au[d][1]),float(au[d][2])])
if(abs(u.length(np.subtract(a,e))-2)<.00001):
thetaz=rv.angle_difference(correct[1],correct[0],e[1],e[0])
thetay=rv.angle_difference(correct[2],correct[0],e[2],e[0])
thetax=rv.angle_difference(correct[2],correct[1],e[2],e[1])
d=-1
for su in range(0,f):
b=np.array([float(s[su+i*f][0]),float(s[su+i*f][1]),float(s[su+i*f][2])])
close_array=np.append(close_array,[np.subtract(rv.move_point(thetax,thetay,thetaz,b),a)],axis=0)
for x in range(0,n):
b=np.array([float(c2[x+n*su+n*f*i][0]),float(c2[x+n*su+n*f*i][1]),float(c2[x+n*su+n*f*i][2])])
close_array=np.append(close_array,[np.subtract(rv.move_point(thetax,thetay,thetaz,b),a)],axis=0)
b=np.array([float(c3[su+i*f][0]),float(c3[su+i*f][1]),float(c3[su+i*f][2])])
close_array=np.append(close_array,[np.subtract(rv.move_point(thetax,thetay,thetaz,b),a)],axis=0)
return close_array
def check_collisions(self):
for player in self.players:
coll = self.track.collide(player.sprite)
if (coll): player.kill()
points = self.track.collide_guides(player.sprite)
player.sprite.intersectpoints = points
self.loop_inputs[player] = [
util.length(player.sprite.rect.center, x) for x in points
]
def compute_length_array(trkfile=None, streams=None, savefname = 'lengths.npy'):
if streams is None and not trkfile is None:
print("Compute length array for fibers in %s" % trkfile)
streams, hdr = tv.read(trkfile, as_generator = True)
n_fibers = hdr['n_count']
if n_fibers == 0:
msg = "Header field n_count of trackfile %s is set to 0. No track seem to exist in this file." % trkfile
print(msg)
raise Exception(msg)
else:
n_fibers = len(streams)
leng = np.zeros(n_fibers, dtype = np.float)
for i,fib in enumerate(streams):
leng[i] = length(fib[0])
# store length array
np.save(savefname, leng)
print("Store lengths array to: %s" % savefname)
return leng
def compute_length_array(trkfile=None, streams=None, savefname='lengths.npy'):
if streams is None and not trkfile is None:
print("Compute length array for fibers in %s" % trkfile)
streams, hdr = tv.read(trkfile, as_generator=True)
n_fibers = hdr['n_count']
if n_fibers == 0:
msg = "Header field n_count of trackfile %s is set to 0. No track seem to exist in this file." % trkfile
print(msg)
raise Exception(msg)
else:
n_fibers = len(streams)
leng = np.zeros(n_fibers, dtype=np.float)
for i, fib in enumerate(streams):
leng[i] = length(fib[0])
# store length array
np.save(savefname, leng)
print("Store lengths array to: %s" % savefname)
return leng
def stft_v2(Buffer, f_n):
w_n = f_n
Buff_len = length(Buffer) # Signal length
# Frequency axis
f_n = math.ceil(f_n / 2) * 2 + 1
Lf = (f_n - 1) / 2
# Time axis
w_n = math.ceil(w_n / 2) * 2 + 1
Lw = (w_n - 1) / 2
# Initialize Spectrum to zero with appropriate size
Spec = np.zeros((f_n, Buff_len), dtype=np.complex64)
## Sliding window over signal
for iterr in range(Buff_len):
i_l = min([iterr, Lw, Lf])
i_r = min([Buff_len - iterr - 1, Lw, Lf])
iter_ind = np.arange(-i_l, i_r + 1)
ind1 = (iter_ind + iterr).astype(
'int') # Time Indexing of the original signal
ind = (iter_ind + Lf).astype('int') # Frequency Indexing of the martix
Spec[ind, iterr] = Buffer[ind1]
# Computing FFT of stacked Windows
Spec = np.fft.fft(Spec.T).T
Spec = Spec * 2 / f_n # normalizing the FFTs
return Spec
def cmat(intrk,
roi_volumes,
parcellation_scheme,
compute_curvature=True,
additional_maps={},
output_types=['gPickle'],
atlas_info={}):
""" Create the connection matrix for each resolution using fibers and ROIs. """
# create the endpoints for each fibers
en_fname = 'endpoints.npy'
en_fnamemm = 'endpointsmm.npy'
#ep_fname = 'lengths.npy'
curv_fname = 'meancurvature.npy'
#intrk = op.join(gconf.get_cmp_fibers(), 'streamline_filtered.trk')
print('Opening file :' + intrk)
fib, hdr = nibabel.trackvis.read(intrk, False)
if parcellation_scheme != "Custom":
resolutions = get_parcellation(parcellation_scheme)
else:
resolutions = atlas_info
# Previously, load_endpoints_from_trk() used the voxel size stored
# in the track hdr to transform the endpoints to ROI voxel space.
# This only works if the ROI voxel size is the same as the DSI/DTI
# voxel size. In the case of DTI, it is not.
# We do, however, assume that all of the ROI images have the same
# voxel size, so this code just loads the first one to determine
# what it should be
firstROIFile = roi_volumes[0]
firstROI = nibabel.load(firstROIFile)
roiVoxelSize = firstROI.get_header().get_zooms()
(endpoints, endpointsmm) = create_endpoints_array(fib, roiVoxelSize, True)
np.save(en_fname, endpoints)
np.save(en_fnamemm, endpointsmm)
# only compute curvature if required
if compute_curvature:
meancurv = compute_curvature_array(fib)
np.save(curv_fname, meancurv)
print("========================")
n = len(fib)
#resolution = gconf.parcellation.keys()
streamline_wrote = False
for parkey, parval in resolutions.items():
#if parval['number_of_regions'] != 83:
# continue
print("Resolution = " + parkey)
# create empty fiber label array
fiberlabels = np.zeros((n, 2))
final_fiberlabels = []
final_fibers_idx = []
# Open the corresponding ROI
print("Open the corresponding ROI")
for vol in roi_volumes:
if parkey in vol:
roi_fname = vol
print roi_fname
#roi_fname = roi_volumes[r]
#r += 1
roi = nibabel.load(roi_fname)
roiData = roi.get_data()
# Create the matrix
nROIs = parval['number_of_regions']
print("Create the connection matrix (%s rois)" % nROIs)
G = nx.Graph()
# add node information from parcellation
gp = nx.read_graphml(parval['node_information_graphml'])
for u, d in gp.nodes_iter(data=True):
G.add_node(int(u), d)
# compute a position for the node based on the mean position of the
# ROI in voxel coordinates (segmentation volume )
G.node[int(u)]['dn_position'] = tuple(
np.mean(np.where(roiData == int(d["dn_correspondence_id"])),
axis=1))
dis = 0
# prepare: compute the measures
t = [c[0] for c in fib]
h = np.array(t, dtype=np.object)
mmap = additional_maps
mmapdata = {}
for k, v in mmap.items():
da = nibabel.load(v)
mmapdata[k] = (da.get_data(), da.get_header().get_zooms())
print("Create the connection matrix")
pc = -1
for i in range(n): # n: number of fibers
# Percent counter
pcN = int(round(float(100 * i) / n))
if pcN > pc and pcN % 1 == 0:
pc = pcN
print('%4.0f%%' % (pc))
# ROI start => ROI end
try:
startROI = int(roiData[
endpoints[i, 0, 0], endpoints[i, 0, 1],
endpoints[i, 0,
2]]) # endpoints from create_endpoints_array
endROI = int(roiData[endpoints[i, 1, 0], endpoints[i, 1, 1],
endpoints[i, 1, 2]])
except IndexError:
print(
"An index error occured for fiber %s. This means that the fiber start or endpoint is outside the volume. Continue."
% i)
continue
# Filter
if startROI == 0 or endROI == 0:
dis += 1
fiberlabels[i, 0] = -1
continue
if startROI > nROIs or endROI > nROIs:
# print("Start or endpoint of fiber terminate in a voxel which is labeled higher")
# print("than is expected by the parcellation node information.")
# print("Start ROI: %i, End ROI: %i" % (startROI, endROI))
# print("This needs bugfixing!")
continue
# Update fiber label
# switch the rois in order to enforce startROI < endROI
if endROI < startROI:
tmp = startROI
startROI = endROI
endROI = tmp
fiberlabels[i, 0] = startROI
fiberlabels[i, 1] = endROI
final_fiberlabels.append([startROI, endROI])
final_fibers_idx.append(i)
# Add edge to graph
if G.has_edge(startROI, endROI):
G.edge[startROI][endROI]['fiblist'].append(i)
else:
G.add_edge(startROI, endROI, fiblist=[i])
print(
"Found %i (%f percent out of %i fibers) fibers that start or terminate in a voxel which is not labeled. (orphans)"
% (dis, dis * 100.0 / n, n))
print("Valid fibers: %i (%f percent)" %
(n - dis, 100 - dis * 100.0 / n))
# create a final fiber length array
finalfiberlength = []
for idx in final_fibers_idx:
# compute length of fiber
finalfiberlength.append(length(fib[idx][0]))
# convert to array
final_fiberlength_array = np.array(finalfiberlength)
# make final fiber labels as array
final_fiberlabels_array = np.array(final_fiberlabels, dtype=np.int32)
# update edges
# measures to add here
for u, v, d in G.edges_iter(data=True):
G.remove_edge(u, v)
di = {
'number_of_fibers': len(d['fiblist']),
}
# additional measures
# compute mean/std of fiber measure
idx = np.where((final_fiberlabels_array[:, 0] == int(u))
& (final_fiberlabels_array[:, 1] == int(v)))[0]
di['fiber_length_mean'] = float(
np.mean(final_fiberlength_array[idx]))
di['fiber_length_std'] = float(np.std(
final_fiberlength_array[idx]))
# this is indexed into the fibers that are valid in the sense of touching start
# and end roi and not going out of the volume
idx_valid = np.where((fiberlabels[:, 0] == int(u))
& (fiberlabels[:, 1] == int(v)))[0]
for k, vv in mmapdata.items():
val = []
for i in idx_valid:
# retrieve indices
try:
idx2 = (h[i] / vv[1]).astype(np.uint32)
val.append(vv[0][idx2[:, 0], idx2[:, 1], idx2[:, 2]])
except IndexError, e:
print "Index error occured when trying extract scalar values for measure", k
print "--> Discard fiber with index", i, "Exception: ", e
print "----"
da = np.concatenate(val)
di[k + '_mean'] = float(da.mean())
di[k + '_std'] = float(da.std())
del da
del val
G.add_edge(u, v, di)
# storing network
if 'gPickle' in output_types:
nx.write_gpickle(G, 'connectome_%s.gpickle' % parkey)
if 'mat' in output_types:
# edges
size_edges = (parval['number_of_regions'],
parval['number_of_regions'])
edge_keys = G.edges(data=True)[0][2].keys()
edge_struct = {}
for edge_key in edge_keys:
edge_struct[edge_key] = nx.to_numpy_matrix(G, weight=edge_key)
# nodes
size_nodes = parval['number_of_regions']
node_keys = G.nodes(data=True)[0][1].keys()
node_struct = {}
for node_key in node_keys:
if node_key == 'dn_position':
node_arr = np.zeros([size_nodes, 3], dtype=np.float)
else:
node_arr = np.zeros(size_nodes, dtype=np.object_)
node_n = 0
for _, node_data in G.nodes(data=True):
node_arr[node_n] = node_data[node_key]
node_n += 1
node_struct[node_key] = node_arr
scipy.io.savemat('connectome_%s.mat' % parkey,
mdict={
'sc': edge_struct,
'nodes': node_struct
})
if 'graphml' in output_types:
g2 = nx.Graph()
for u_gml, v_gml, d_gml in G.edges_iter(data=True):
g2.add_edge(u_gml, v_gml, d_gml)
for u_gml, d_gml in G.nodes(data=True):
g2.add_node(
u_gml, {
'dn_correspondence_id': d_gml['dn_correspondence_id'],
'dn_fsname': d_gml['dn_fsname'],
'dn_hemisphere': d_gml['dn_hemisphere'],
'dn_name': d_gml['dn_name'],
'dn_position_x': float(d_gml['dn_position'][0]),
'dn_position_y': float(d_gml['dn_position'][1]),
'dn_position_z': float(d_gml['dn_position'][2]),
'dn_region': d_gml['dn_region']
})
nx.write_graphml(g2, 'connectome_%s.graphml' % parkey)
print("Storing final fiber length array")
fiberlabels_fname = 'final_fiberslength_%s.npy' % str(parkey)
np.save(fiberlabels_fname, final_fiberlength_array)
print("Storing all fiber labels (with orphans)")
fiberlabels_fname = 'filtered_fiberslabel_%s.npy' % str(parkey)
np.save(
fiberlabels_fname,
np.array(fiberlabels, dtype=np.int32),
)
print("Storing final fiber labels (no orphans)")
fiberlabels_noorphans_fname = 'final_fiberlabels_%s.npy' % str(parkey)
np.save(fiberlabels_noorphans_fname, final_fiberlabels_array)
if not streamline_wrote:
print("Filtering tractography - keeping only no orphan fibers")
finalfibers_fname = 'streamline_final.trk'
save_fibers(hdr, fib, finalfibers_fname, final_fibers_idx)
def active_sess_dechirp(x_1):
#ACTIVE_SESS_DECHIRP Summary of this function goes here
# Detailed explanation goes here
SF = param_configs(1)
BW = param_configs(2)
Fs = param_configs(3)
N = int(2**SF)
upsampling_factor = int(Fs / BW)
DC = np.conj(sym_to_data_ang([1], N))
DC_fft = np.fft.fft(DC)
DC_upsamp = np.fft.ifft(
np.concatenate([
DC_fft[:N // 2 + 1],
np.zeros((upsampling_factor - 1) * N), DC_fft[N // 2 + 1:N]
]))
peak_gain = []
uplink_wind = []
n = []
p = []
last_wind = 0
win_jump_factor = 3
front_buf = 6 * win_jump_factor
back_buf = 3 * win_jump_factor
win_jump = math.floor(N * upsampling_factor / win_jump_factor)
mov_thresh_wind = 1000 * win_jump_factor
mov_thresh = 0
mov_thresh_rec = []
for i in tqdm(range(math.floor(length(x_1) / win_jump) - win_jump_factor)):
wind = x_1[i * win_jump:i * win_jump + (N * upsampling_factor)]
wind_fft = np.abs(np.fft.fft(wind * DC_upsamp)) # 17 sec
wind_fft = wind_fft[np.concatenate([np.arange(N//2, dtype=int), \
np.arange((N//2 + (upsampling_factor-1)*N), (upsampling_factor)*N, dtype=int)])]
noise_floor = np.mean(wind_fft) # 4 sec
n.append(noise_floor)
fft_peak = wind_fft.max(0)
p.append(fft_peak)
peak_gain.append(10 * math.log10(fft_peak / noise_floor))
if i + 1 > mov_thresh_wind:
mov_thresh = 1.3 * np.mean(peak_gain[-mov_thresh_wind + 1:])
if mov_thresh > 6:
mov_thresh = 6
else:
mov_thresh = 1.3 * np.mean(peak_gain)
if mov_thresh > 6:
mov_thresh = 6
mov_thresh_rec.append(mov_thresh)
if peak_gain[-1] >= mov_thresh:
if i + 1 > last_wind:
if i - back_buf < 1:
uplink_wind.append([1, i + 1 + front_buf])
else:
uplink_wind.append([i + 1 - back_buf, i + 1 + front_buf])
last_wind = uplink_wind[-1][1]
elif i + 1 <= last_wind:
uplink_wind[-1][1] = i + 1 + front_buf
last_wind = uplink_wind[-1][1]
uplink_wind = np.array(uplink_wind)
uplink_wind = uplink_wind[((uplink_wind[:, 1] - uplink_wind[:, 0]) !=
(front_buf + back_buf)).nonzero()[0], :]
uplink_wind = uplink_wind[((uplink_wind[:, 1] - uplink_wind[:, 0]) !=
(front_buf + back_buf + 1)).nonzero()[0], :]
uplink_wind = uplink_wind[((uplink_wind[:, 1] - uplink_wind[:, 0]) !=
(front_buf + back_buf - 1)).nonzero()[0], :]
temp_link = uplink_wind
uplink_wind = uplink_wind * win_jump
if uplink_wind[-1, 1] > length(x_1):
uplink_wind[-1, 1] = length(x_1)
return uplink_wind
def dnsamp_buff(Data_stack,Upchirp_ind):
# load Parameters
SF = param_configs(1)
BW = param_configs(2)
Fs = param_configs(3)
N = int(2**SF)
num_preamble = param_configs(4)
num_sync = param_configs(5)
num_DC = param_configs(6)
DC = np.conj(sym_to_data_ang([1],N))
####################################
## Compute and Correct Frequency Offsets for each Preamble Detected in each Data_stack and Find the Peak Statistics needed for demodulation
Up_ind = []
peak_amp = []
Data_buff = []
ffo = []
FFO = []
# n_pnt is the fft Factor - fft(Signal, n_pnt * length(Signal))
n_pnt = 16
peak_stats = []
# iterate over all Upchirps that qualified 8 consecutive Peak condition
for k in range(Upchirp_ind.shape[0]):
if(Upchirp_ind[k,0] - N <= 0):
peak_stats.append([])
continue
inn = []
k_peak_stats = []
Data_freq_off = []
# iterate overall downsampled buffers
for m in range(Data_stack.shape[0]):
data_wind = []
data_fft = []
freq_off = []
# ind_temp contains the Frequency Bins around bin 1 where a
# Preamble Peak can lie
ind_temp = np.concatenate([np.arange(5*n_pnt), np.arange((N*n_pnt)-(4*n_pnt)-1, (N*n_pnt))])
# iterate over all Preambles
c = []
for j in range(num_preamble):
data_wind = Data_stack[m,int(Upchirp_ind[k,0]) - 1 : int(Upchirp_ind[k,0] + (num_preamble*N) -1)]
data_fft.append(abs(np.fft.fft(data_wind[((j)*N):((j+1)*N)] * DC[:N],n_pnt*N)))
c.append(data_fft[j][ind_temp].argmax(0))
c[j] = ind_temp[c[j]] + 1
# Handle -ve and +ve Frequency Offsets Accordingly
if(c[j] > (n_pnt*N)/2):
freq_off.append(( (N*n_pnt) - c[j] ) / n_pnt)
else:
freq_off.append(-1*( c[j] - 1 ) / n_pnt)
# average the frequency offset of 6 middle Preambles
freq_off = np.sum( freq_off[1:7] ) / (num_preamble - 2)
ffo.append(freq_off)
# Correct for the Frequency Offset in corresponding Data_Stack
Data_freq_off.append(Data_stack[m,:] * np.exp( (1j*2*math.pi*(freq_off / N)) * np.arange(1, length(Data_stack[m,:]) + 1) ))
# ind_temp contains the Frequency Bins around bin 1 where a
# Preamble Peak can lie, assumption (-5*BW/2^SF <= Freq_off <= 5*BW/2^SF)
ind_temp = np.concatenate([range(5), range(N-4, N)])
a = []
c = []
data_wind = []
data_fft = []
# for the frequency offset corrected Data Stack, find FFT of Preamble to get Peak Statistics
for j in range(num_preamble):
data_wind = Data_freq_off[m][int(Upchirp_ind[k,0]) - 1 : int(Upchirp_ind[k,0] + (num_preamble*N)) - 1]
data_fft.append(abs(np.fft.fft(data_wind[(j)*N : (j+1)*N] * DC[:N],N)))
[aj,cj] = data_fft[j][ind_temp].max(0), data_fft[j][ind_temp].argmax(0)
a.append(aj); c.append(cj)
c[j] = ind_temp[c[j]]
k_peak_stats.append([np.mean(a), np.var(a, ddof=1), np.std(a, ddof=1)])
## Find the Right Data_stack to work with
# first find the stft of given stack at the Preamble Region,
# Spec is a 2D Matrix, rows - Freq. Bins & Col. - Time Samples
Spec = stft_v1(Data_freq_off[m][int(Upchirp_ind[k,0] - N)-1:int(Upchirp_ind[k,-1] + N - 1 - N)],N,DC[:N],0,0)
temp = []
freq_track_qual = []
pream_peak_ind = []
adj_ind = []
# row_ind contains the Frequency Rows around bin 1 where a
# Preamble Peak can lie
row_ind = np.concatenate([range(N-6,N), range(0,6)])
count = 1
for i in np.nditer(row_ind):
temp.append(np.sum(np.abs(Spec[i,:])))
count = count + 1
temp = np.array(temp)
# Frequency Track in row containing Preamble should have
# maximum energy
ind = temp.argmax(0)
pream_peak_ind = row_ind[ind]
# Find row indices for Preamble row + 1 & - 1
adj_ind = np.array([np.mod(pream_peak_ind-1+1,N), np.mod(pream_peak_ind+1+1,N)]) # plus 1 for index conversion
if(np.sum(adj_ind == 0) == 1):
adj_ind[(adj_ind == 0).nonzero()] = N
# A good quality frequency track for a preamble is one that has
# least energy leakage in adjacent rows (this promises very sharp FFT peaks)
adj_ind -= 1 # subtract 1 to convert back to Python indices
freq_track_qual = ( np.sum(np.abs(Spec[pream_peak_ind,:])) - np.sum(np.abs(Spec[adj_ind[0],:])) ) + ( np.sum(np.abs(Spec[pream_peak_ind,:])) - np.sum(np.abs(Spec[adj_ind[1],:])) )
inn.append(freq_track_qual)
inn = np.array(inn)
peak_stats.append(k_peak_stats)
Data_freq_off = np.array(Data_freq_off)
# choosing the best Data_stack based on maximum energy difference from
# adjacent bins
b = inn.argmax(0)
# output frequency offset corrected buffer with relevant, Peak
# statistics and frequency offsets
Data_buff.append(Data_freq_off[b,:])
FFO.append(ffo[b])
peak_amp.append(peak_stats[k][b])
Up_ind.append(Upchirp_ind[k,:])
Data_buff = np.array(Data_buff)
Up_ind = np.array(Up_ind)
peak_amp = np.array(peak_amp)
return [Data_buff, peak_amp, Up_ind, FFO]
def visible(self, eye, point):
v = util.sub(eye, point)
o = point
d = util.normalize(v)
t = self.intersect(o, d, 0, util.length(v))
return t is None
def draw_shaded(loops, base_colour, rim_colour, distance):
gl.glDepthFunc(gl.GL_ALWAYS)
gl.glColor3fv(base_colour)
gl.glBegin(gl.GL_TRIANGLES)
for tri in util.triangulate(loops):
for point in tri:
gl.glVertex3f(*point, -1)
gl.glEnd()
gl.glDepthFunc(gl.GL_LEQUAL)
gl.glBegin(gl.GL_TRIANGLES)
for loop in loops:
for a, b, c in zip(np.roll(loop, -1, 0), loop, np.roll(loop, 1, 0)):
normal = np.array([b[1] - a[1], a[0] - b[0]])
l = util.length(normal)
if l == 0:
continue
normal /= l
inset_a = a - normal * distance
inset_b = b - normal * distance
gl.glColor3fv(rim_colour)
gl.glVertex3f(*a, 0)
gl.glVertex3f(*b, 0)
gl.glColor3fv(base_colour)
gl.glVertex3f(*inset_a, -1)
gl.glVertex3f(*inset_a, -1)
gl.glVertex3f(*inset_b, -1)
gl.glColor3fv(rim_colour)
gl.glVertex3f(*b, 0)
gl.glEnd()
for loop in loops:
for a, b, c in zip(np.roll(loop, -1, 0), loop, np.roll(loop, 1, 0)):
if util.is_convex(a, b, c):
continue
normals = [
np.array([b[1] - a[1], a[0] - b[0]]),
np.array([c[1] - b[1], b[0] - c[0]]),
]
normals = [norm / util.length(norm) for norm in normals]
for _ in range(2):
new_normals = []
for normA, normB in zip(normals, normals[1:]):
new = normA + normB
new /= util.length(new)
new_normals.append(new)
for i, norm in enumerate(new_normals):
normals.insert(1 + 2 * i, norm)
gl.glBegin(gl.GL_TRIANGLE_FAN)
gl.glColor3fv(rim_colour)
gl.glVertex3f(*b, 0)
gl.glColor3fv(base_colour)
for norm in normals:
point = b - norm * distance
gl.glVertex3f(*point, -1)
gl.glEnd()
gl.glColor3f(0, 0, 0)
for loop in loops:
gl.glBegin(gl.GL_LINE_LOOP)
for point in loop:
gl.glVertex3f(*point, 1)
gl.glEnd()
# Find the vertices of the bounding box from the lines
v = (u.intersection(le[0], se[0]),
u.intersection(le[0], se[1]),
u.intersection(le[1], se[0]),
u.intersection(le[1], se[1]))
# Show the raw vertices without epsilon
if SHOW:
plt.scatter(np.array(v)[:,0], np.array(v)[:,1])
# Calculate the diagonal lines of the bounding box
diags = (u.line(v[0], v[3]), u.line(v[1], v[2]))
# Calculate the center, width, height, and angle of the bounding box
c = u.intersection(diags[0], diags[1]) # Find the center by looking at the intersection
w = u.length(v[0], v[1]) + EPSILON # Find the width
h = u.length(v[0], v[2]) + EPSILON # Find the height
a = u.angle(v[3], v[2]) # Find the angle
wBucket = u.bucketCount(wBucket, w, 10)
hBucket = u.bucketCount(hBucket, h, 10)
aBucket = u.bucketCount(aBucket, a, 10)
if SHOW:
plt.scatter(c[0], c[1])
plt.annotate(a, (c[0], c[1]))
line = str(c[0]) + " " + str(c[1]) + " " + str(w) + " " + str(h) + " " + str(label) + " " + str(a) + "\n"
rboxLines.append(line)
if PRINT:
if os.path.exists(symbols_ground_truth_path):
sym = np.loadtxt(symbols_ground_truth_path)
calculate_SER = True
else:
calculate_SER = False
# Generating a Downchirp
DC = np.conj(sym_to_data_ang([1],N))
# parse complex data
x_1 = np.fromfile(in_file_path, dtype=np.complex64)
# file duration
t = np.arange(len(x_1)) / Fs
x_1 = x_1[:int(math.floor(length(x_1) / upsampling_factor) * upsampling_factor)]
x_1_dnsamp = x_1[::int(upsampling_factor)]
file_dur = length(x_1) / Fs
print('Active session dechirping:')
uplink_wind = active_sess_dechirp(x_1) - 1 # subtract 1 for indexing
print('Detected ', len(uplink_wind), ' active sessions')
demod_sym_stack = []
Peaks = []
for m in range(uplink_wind.shape[0]):
print('Packet ' + str(m+1) + '/' + str(uplink_wind.shape[0]))
# DC correlations to find LoRa pkts out of collision
temp_buff = []
temp_buff = x_1[int(uplink_wind[m, 0]) : int(uplink_wind[m, 1]) + 1]
def four_body_func(vecs, t, *args):
pos_1 = vecs[0:2]
pos_2 = vecs[2:4]
pos_3 = vecs[4:6]
pos_4 = vecs[6:8]
vel_1 = vecs[8:10]
vel_2 = vecs[10:12]
vel_3 = vecs[12:14]
vel_4 = vecs[14:16]
acc_1_2 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_2, pos_1)), [args[1]] ),
math.pow(u.length(u.position_sub(pos_2, pos_1)), 3))
acc_2_1 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_1, pos_2)), [args[0]] ),
math.pow(u.length(u.position_sub(pos_2, pos_1)), 3))
acc_3_1 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_1, pos_3)), [args[0]] ),
math.pow(u.length(u.position_sub(pos_3, pos_1)), 3))
acc_1_3 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_3, pos_1)), [args[2]] ),
math.pow(u.length(u.position_sub(pos_3, pos_1)), 3))
acc_4_1 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_1, pos_4)), [args[0]] ),
math.pow(u.length(u.position_sub(pos_4, pos_1)), 3))
acc_1_4 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_4, pos_1)), [args[3]] ),
math.pow(u.length(u.position_sub(pos_4, pos_1)), 3))
acc_3_2 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_2, pos_3)), [args[1]] ),
math.pow(u.length(u.position_sub(pos_3, pos_2)), 3))
acc_2_3 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_3, pos_2)), [args[2]] ),
math.pow(u.length(u.position_sub(pos_3, pos_2)), 3))
acc_4_2 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_2, pos_4)), [args[1]] ),
math.pow(u.length(u.position_sub(pos_4, pos_2)), 3))
acc_2_4 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_4, pos_2)), [args[3]] ),
math.pow(u.length(u.position_sub(pos_4, pos_2)), 3))
acc_3_4 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_4, pos_3)), [args[3]] ),
math.pow(u.length(u.position_sub(pos_4, pos_3)), 3))
acc_4_3 = u.position_div_scalars(u.position_mult_scalars( (u.position_sub(pos_3, pos_4)), [args[3]] ),
math.pow(u.length(u.position_sub(pos_4, pos_3)), 3))
acc_1 = u.sum_vectors([acc_1_2, acc_1_3, acc_1_4])
acc_2 = u.sum_vectors([acc_2_1, acc_2_3, acc_2_4])
acc_3 = u.sum_vectors([acc_3_1, acc_3_2, acc_3_4])
acc_4 = u.sum_vectors([acc_4_1, acc_4_2, acc_4_3])
result = np.array(u.flatten([vel_1.tolist(), vel_2.tolist(), vel_3.tolist(), vel_4.tolist(),
acc_1, acc_2, acc_3, acc_4]))
#print "Result of func: ", result
return result
chosen_marker = None
for marker in sorted_markers:
next_footprint_tf = marker[1].drift_cor_data[0]
previous_footprint_tf = self.current_map_to_footprint_tf
#first marker caugth ever
if previous_footprint_tf == None:
previous_footprint_tf = next_footprint_tf
change_tf = np.dot(next_footprint_tf, tfmath.inverse_matrix(previous_footprint_tf))
change_tf_tran, change_tf_rot = util.tran_and_euler_from_matrix(change_tf)
time_diff = (rospy.Time.now() - self.last_update_timestamp).to_sec()
distance = util.length(change_tf_tran)
angle_diff = change_tf_rot[2]
rospy.loginfo("Looking at marker %s", marker[0])
rospy.loginfo("\t{0:.3f} m, {1:.3f} radians".format(distance, angle_diff))
'''
lin_velocity = distance / time_diff
ang_velocity = angle_diff / time_diff
rospy.loginfo("\t{0:.3f} > {1:.3f}, diff: {2:.3f}".format(lin_velocity, self.max_lin_vel, lin_velocity-self.max_lin_vel))
rospy.loginfo("\t{0:.3f} > {1:.3f}, diff: {2:.3f}".format(ang_velocity, self.max_ang_vel, ang_velocity-self.max_ang_vel))
'''
if time_diff < self.test_time_diff:
rospy.loginfo("Time check less than test time, checking vels")
def angle(v1, v2, v3, inputfile):
vector1 = u.part_vector(v2, v1, inputfile)
vector2 = u.part_vector(v2, v3, inputfile)
angle = np.arccos(
np.dot(vector1, vector2) / (u.length(vector1) * u.length(vector2)))
return np.degrees(angle)
def CIC_Demod(Pream_ind,Rx_Buffer,Pream_ind_stack,Peak_amp,m):
#DEMOD
# CIC Demodulation
# chirp variables
SF = param_configs(1)
N = int(2**SF)
# LORA pkt variables
num_preamble = param_configs(4)
num_sync = param_configs(5)
num_DC = param_configs(6)
num_data_sym = param_configs(7)
DC = np.conj(sym_to_data_ang([1], N))
###########################################################################
Pream_ind_stack = np.append(Pream_ind_stack, Pream_ind_stack[:,num_preamble-1] + N)
###########################################################################
# for each Preamble in the Pream_ind_stack, compute exact start and end
# indices for all the data symbols
frm_ind = []
for i in range(Pream_ind_stack.shape[0]):
frm_st = Pream_ind_stack[i] + (num_preamble*N) + (num_DC*N) + (num_sync*N)
frm_ind.append([np.arange(frm_st, frm_st+((num_data_sym-1)*N)+1, N), \
np.arange(frm_st+N-1, frm_st+((num_data_sym)*N)+1, N)])
frm_ind = np.array(frm_ind) - 1 # subtract 1 for index conversion
# for the pkt to be demodulated, find indices for each data symbol
data_ind = []
en_arr = []
sym_peak = []
Data_frame_start = Pream_ind[0] + (num_preamble*N) + (num_DC*N) + (num_sync*N)
Data_frame_end = Data_frame_start + (num_data_sym*N)
frame_indices = np.array([(np.arange(Data_frame_start, Data_frame_start+((num_data_sym-1)*N)+1, N)),
(np.arange(Data_frame_start+N-1, Data_frame_start+((num_data_sym)*N)+1, N))]).T.astype('int') - 1 # subtract 1 for zero indexing
data_ind = []
sym_peak = []
symbols = []
for k in range(num_data_sym):
## Find interfering Symbol Boundaries
# for current demodulation window, find the chunks of interfering
# symbols due to collisions by determining index overlaps
ind = []
sym_bnd = []
for i in range(len(frm_ind)):
if i != m:
st = frm_ind[i][0]
ed = frm_ind[i][1]
newst = st[np.intersect1d((st > frame_indices[k,0]).nonzero() , (st < frame_indices[k,1]).nonzero())]
newed = ed[np.intersect1d((ed > frame_indices[k,0]).nonzero() , (ed < frame_indices[k,1]).nonzero())]
if len(newst) != 0:
sym_bnd.append(newst)
if len(newed) != 0:
sym_bnd.append(newed)
sym_bnd = np.array(sym_bnd)
## CIC Filtering
if frame_indices[k, 1] >= len(Rx_Buffer):
symbols.append(-3)
continue
# standard LoRa dechirping
data_wind = Rx_Buffer[frame_indices[k,0]:frame_indices[k,1]+1] * DC
data_fft = np.abs(np.fft.fft(data_wind))
# scale the dmeodulation window with appropriate Gaussian to
# suppress the interfering symbols at window ends
sigma = 1
WinFun = np.exp(-(1/(2*(sigma**2)))* np.linspace(-1,1,N) ** 2)
WinFun = WinFun / (math.sqrt(2*math.pi)*sigma)
temp_wind = data_wind * WinFun
sym_bnd = np.mod(sym_bnd - frame_indices[k,0],N)
intf_wind = []
# n_pnt is the fft Factor - fft(Signal, n_pnt * length(Signal))
n_pnt = 4
for i in range(len(sym_bnd)):
buff = np.zeros((2,n_pnt*N), dtype=np.complex64)
sym_bnd_i = int(sym_bnd[np.unravel_index(i, sym_bnd.shape, 'F')])
buff[0,:sym_bnd_i - 1] = temp_wind[:sym_bnd_i - 1]
buff[0,:] = np.abs(np.fft.fft(buff[0,:], n_pnt*N)) / np.sqrt(np.sum(np.abs(buff[0,:]) ** 2)) # div OKAY
buff[1,sym_bnd_i-1:N] = temp_wind[sym_bnd_i-1:N]
buff[1,:] = np.abs(np.fft.fft(buff[1,:],n_pnt*N)) / np.sqrt(np.sum(np.abs(buff[1,:]) ** 2))
intf_wind = buff
# CIC's min operation to suppress interfering symbol's Peaks and
# then finding candidate symbols
intf_wind = np.array(intf_wind)
intf_wind_min_fft = np.amin(intf_wind,axis=0) if len(intf_wind) != 0 else []
pot_sym_cic = get_max(intf_wind_min_fft,4*np.sum(intf_wind_min_fft)/(n_pnt*N),n_pnt*N)
pot_sym_cic = np.ceil(pot_sym_cic/n_pnt)
# Out of all peaks, find ones with in range of +- 0.5*Preamble Peak
PwrFctr = 0.5
PwrFlr = 4 # may try 3
up_thresh = (Peak_amp[0] + PwrFctr*Peak_amp[0])
low_thresh = (Peak_amp[0] - PwrFctr*Peak_amp[0])
if(low_thresh < (PwrFlr*np.sum(data_fft)/N)): #1
low_thresh = (PwrFlr*np.sum(data_fft)/N)
pot_sym_pf = get_bounded_max(data_fft,up_thresh,low_thresh)
## Filtering Preamble of interfering Packets
# Filter out Peaks with in current window that are appearinf repeatedly
# in 3 consecutive windows
if not (frame_indices[k,1] + N > len(Rx_Buffer) or frame_indices[k,1] + 2*N > len(Rx_Buffer)):
data_wind_next_1 = Rx_Buffer[frame_indices[k,0] + N : frame_indices[k,1] + N+1] * DC
data_wind_prev_1 = Rx_Buffer[frame_indices[k,0] - N : frame_indices[k,1] - N+1] * DC
data_wind_next_2 = Rx_Buffer[frame_indices[k,0] + 2*N : frame_indices[k,1] + 2*N+1] * DC
data_wind_prev_2 = Rx_Buffer[frame_indices[k,0] - 2*N : frame_indices[k,1] - 2*N+1] * DC
temp_next_1 = np.abs(np.fft.fft(data_wind_next_1,N))
temp_prev_1 = np.abs(np.fft.fft(data_wind_prev_1,N))
temp_next_2 = np.abs(np.fft.fft(data_wind_next_2,N))
temp_prev_2 = np.abs(np.fft.fft(data_wind_prev_2,N))
next_wind_sym_1 = get_bounded_max(temp_next_1,4*np.sum(temp_next_1)/N,N) # TODO changed these four lines
next_wind_sym_2 = get_bounded_max(temp_next_2,4*np.sum(temp_next_2)/N,N)
prev_wind_sym_1 = get_bounded_max(temp_prev_1,4*np.sum(temp_prev_1)/N,N)
prev_wind_sym_2 = get_bounded_max(temp_prev_2,4*np.sum(temp_prev_2)/N,N)
temp = []
for i in range(length(pot_sym_pf)):
if( (np.sum(pot_sym_pf[i] == prev_wind_sym_1) and np.sum(pot_sym_pf[i] == next_wind_sym_1))\
or (np.sum(pot_sym_pf[i] == prev_wind_sym_2) and np.sum(pot_sym_pf[i] == prev_wind_sym_1))\
or (np.sum(pot_sym_pf[i] == next_wind_sym_1) and np.sum(pot_sym_pf[i] == next_wind_sym_2)) ):
pass
else:
temp.append(pot_sym_pf[i])
pot_sym_pf = np.array(temp) + 1 # convert from indices to data
## Freq. Offset Filtering
# since we have removed Frequency Offset from pkt under consideration 'Pream_ind'
# and chosen the right downsampled buffer, the true symbol peak should
# be the most crisp one (either use this or Choir Module(next), results should be almost similar)
# and the interfering symbols peak may or may not be crisp
temp = []
for i in range(length(pot_sym_pf)):
if sum(pot_sym_pf[i] + 1 == pot_sym_pf) or sum(pot_sym_pf[i] - 1 == pot_sym_pf): #TODO what is going on here???
pass
else:
temp.append(pot_sym_pf[i])
pot_sym = np.array(temp)
## Choir Module
# npnt = 16
# data_fft_npnt = abs(np.fft.fft(data_wind,npnt*N))
# FO_thresh = 0.25
# sym_FO = []
# temp = []
# for i in range(length(pot_sym_pf)):
# ind = []
# if(pot_sym_pf[i] == 1):
# ind = np.concatenate([np.arange((N*npnt) - (npnt/2) + 1, (N*npnt)), np.arange((((pot_sym_pf[i]-1) * npnt) + 1) + (npnt/2), N*npnt)])
# else:
# ind = np.arange((((pot_sym_pf[i]-1) * npnt) + 1) - (npnt/2), (((pot_sym_pf[i]-1) * npnt) + 1) + (npnt/2) + 1)
# # ind OKAY
# ind = ind.astype('int') - 1 # convert back from data to indices
# a = np.argmax(data_fft_npnt[ind])
# sym_FO.append(abs(a - ((npnt/2)+1))/npnt)
# if(sym_FO[-1] < FO_thresh):
# temp.append(pot_sym_pf[i])
# pot_sym = np.array(temp)
#################################################################
## Make the final decision
b = []
if(length(sym_bnd) == 0):
# if there is no symbol colliding with current demod window
if(len(pot_sym) == 0):
symbols.append(np.argmax(data_fft) + 1)
else:
# choose peak closest in height to Preamble Peak
dist = abs(data_fft[pot_sym - 1] - (up_thresh + low_thresh)/2)
b = np.argmin(dist)
symbols.append(pot_sym[b])
else:
# if symbols are colliding with current demod window
fin_sym = (np.intersect1d(pot_sym_cic,pot_sym)).astype('int')
## Final Decision
if(length(fin_sym) == 0):
if length(pot_sym_cic) == 0 and length(pot_sym) != 0:
# choose peak closest in height to Preamble Peak
dist = abs(data_fft[pot_sym - 1] - (up_thresh + low_thresh)/2)
b = np.argmin(dist)
symbols.append(pot_sym[b])
elif length(pot_sym) == 0 and length(pot_sym_cic) != 0:
# make decision based on CIC's windows, correct symbol should
# have lowest std as it appears in all windows
sdev = np.std(intf_wind[:,(n_pnt * pot_sym_cic).astype('int')])
b = np.argmin(sdev)
symbols.append(pot_sym_cic[b])
elif length(pot_sym) == 0 and length(pot_sym_cic) == 0:
symbols.append(np.argmax(data_fft) + 1) # add one to account for zero-indexing
else:
# choose peak closest in height to Preamble Peak
dist = abs(data_fft[pot_sym - 1] - (up_thresh + low_thresh)/2)
b = np.argmin(dist)
symbols.append(pot_sym[b])
else:
# if intersection yields some candidates then decide based on partial STFT as following
## Stft
# find stft 2D matrix of follwoing dimensions
# N frequency (rows) x [1 : avg_pnts N - avg_pnts : N] (columns)
# i.e. finding Spectrum of first 10 and last 10 time samples (Dont need to compute whole Spectrum)
avg_pnts = 10 # number of start and end time samples to average over
G_wind1 = data_wind
Spec = np.zeros((N, 2*(avg_pnts+1)), dtype=np.complex64)
for i in range(avg_pnts+1):
Spec[:int(N//2) + i,i] = G_wind1[:int(N//2) + i]
Spec[int(N//2) - (avg_pnts - i)-1:N,i+1+avg_pnts] = G_wind1[int(N//2) - (avg_pnts - i)-1:N]
Spec = np.fft.fft(Spec, axis=0)
# the amplitude difference at the start and end of the spectrum
# for correct symbol should be minimum, (non-interfering symbol's
# frequency track appears continuosly for all the columns)
spec_slice1 = np.abs(Spec[fin_sym-1,:(avg_pnts+1)]) # TODO why minus 2??
freq_amp = np.amin(spec_slice1,axis=1) if len(spec_slice1) != 0 else []
spec_slice2 = np.abs(Spec[fin_sym-1,-avg_pnts-1:])
freq_amp_end = np.amin(spec_slice2,axis=1) if len(spec_slice2) != 0 else []
dif = np.abs(freq_amp - freq_amp_end)
b = np.argmin(dif)
symbols.append(fin_sym[b])
return symbols
def cmat(intrk, roi_volumes, parcellation_scheme, compute_curvature=True, additional_maps={}, output_types=['gPickle'], atlas_info = {}):
""" Create the connection matrix for each resolution using fibers and ROIs. """
# create the endpoints for each fibers
en_fname = 'endpoints.npy'
en_fnamemm = 'endpointsmm.npy'
#ep_fname = 'lengths.npy'
curv_fname = 'meancurvature.npy'
#intrk = op.join(gconf.get_cmp_fibers(), 'streamline_filtered.trk')
print('Opening file :' + intrk)
fib, hdr = nibabel.trackvis.read(intrk, False)
if parcellation_scheme != "Custom":
resolutions = get_parcellation(parcellation_scheme)
else:
resolutions = atlas_info
# Previously, load_endpoints_from_trk() used the voxel size stored
# in the track hdr to transform the endpoints to ROI voxel space.
# This only works if the ROI voxel size is the same as the DSI/DTI
# voxel size. In the case of DTI, it is not.
# We do, however, assume that all of the ROI images have the same
# voxel size, so this code just loads the first one to determine
# what it should be
firstROIFile = roi_volumes[0]
firstROI = nibabel.load(firstROIFile)
roiVoxelSize = firstROI.get_header().get_zooms()
(endpoints,endpointsmm) = create_endpoints_array(fib, roiVoxelSize, True)
np.save(en_fname, endpoints)
np.save(en_fnamemm, endpointsmm)
# only compute curvature if required
if compute_curvature:
meancurv = compute_curvature_array(fib)
np.save(curv_fname, meancurv)
print("========================")
n = len(fib)
#resolution = gconf.parcellation.keys()
streamline_wrote = False
for parkey, parval in resolutions.items():
#if parval['number_of_regions'] != 83:
# continue
print("Resolution = "+parkey)
# create empty fiber label array
fiberlabels = np.zeros( (n, 2) )
final_fiberlabels = []
final_fibers_idx = []
# Open the corresponding ROI
print("Open the corresponding ROI")
for vol in roi_volumes:
if parkey in vol:
roi_fname = vol
print roi_fname
#roi_fname = roi_volumes[r]
#r += 1
roi = nibabel.load(roi_fname)
roiData = roi.get_data()
# Create the matrix
nROIs = parval['number_of_regions']
print("Create the connection matrix (%s rois)" % nROIs)
G = nx.Graph()
# add node information from parcellation
gp = nx.read_graphml(parval['node_information_graphml'])
for u,d in gp.nodes_iter(data=True):
G.add_node(int(u), d)
# compute a position for the node based on the mean position of the
# ROI in voxel coordinates (segmentation volume )
G.node[int(u)]['dn_position'] = tuple(np.mean( np.where(roiData== int(d["dn_correspondence_id"]) ) , axis = 1))
dis = 0
# prepare: compute the measures
t = [c[0] for c in fib]
h = np.array(t, dtype = np.object )
mmap = additional_maps
mmapdata = {}
for k,v in mmap.items():
da = nibabel.load(v)
mmapdata[k] = (da.get_data(), da.get_header().get_zooms() )
print("Create the connection matrix")
pc = -1
for i in range(n): # n: number of fibers
# Percent counter
pcN = int(round( float(100*i)/n ))
if pcN > pc and pcN%1 == 0:
pc = pcN
print('%4.0f%%' % (pc))
# ROI start => ROI end
try:
startROI = int(roiData[endpoints[i, 0, 0], endpoints[i, 0, 1], endpoints[i, 0, 2]]) # endpoints from create_endpoints_array
endROI = int(roiData[endpoints[i, 1, 0], endpoints[i, 1, 1], endpoints[i, 1, 2]])
except IndexError:
print("An index error occured for fiber %s. This means that the fiber start or endpoint is outside the volume. Continue." % i)
continue
# Filter
if startROI == 0 or endROI == 0:
dis += 1
fiberlabels[i,0] = -1
continue
if startROI > nROIs or endROI > nROIs:
# print("Start or endpoint of fiber terminate in a voxel which is labeled higher")
# print("than is expected by the parcellation node information.")
# print("Start ROI: %i, End ROI: %i" % (startROI, endROI))
# print("This needs bugfixing!")
continue
# Update fiber label
# switch the rois in order to enforce startROI < endROI
if endROI < startROI:
tmp = startROI
startROI = endROI
endROI = tmp
fiberlabels[i,0] = startROI
fiberlabels[i,1] = endROI
final_fiberlabels.append( [ startROI, endROI ] )
final_fibers_idx.append(i)
# Add edge to graph
if G.has_edge(startROI, endROI):
G.edge[startROI][endROI]['fiblist'].append(i)
else:
G.add_edge(startROI, endROI, fiblist = [i])
print("Found %i (%f percent out of %i fibers) fibers that start or terminate in a voxel which is not labeled. (orphans)" % (dis, dis*100.0/n, n) )
print("Valid fibers: %i (%f percent)" % (n-dis, 100 - dis*100.0/n) )
# create a final fiber length array
finalfiberlength = []
for idx in final_fibers_idx:
# compute length of fiber
finalfiberlength.append( length(fib[idx][0]) )
# convert to array
final_fiberlength_array = np.array( finalfiberlength )
# make final fiber labels as array
final_fiberlabels_array = np.array(final_fiberlabels, dtype = np.int32)
# update edges
# measures to add here
for u,v,d in G.edges_iter(data=True):
G.remove_edge(u,v)
di = { 'number_of_fibers' : len(d['fiblist']), }
# additional measures
# compute mean/std of fiber measure
idx = np.where( (final_fiberlabels_array[:,0] == int(u)) & (final_fiberlabels_array[:,1] == int(v)) )[0]
di['fiber_length_mean'] = float( np.mean(final_fiberlength_array[idx]) )
di['fiber_length_std'] = float( np.std(final_fiberlength_array[idx]) )
# this is indexed into the fibers that are valid in the sense of touching start
# and end roi and not going out of the volume
idx_valid = np.where( (fiberlabels[:,0] == int(u)) & (fiberlabels[:,1] == int(v)) )[0]
for k,vv in mmapdata.items():
val = []
for i in idx_valid:
# retrieve indices
try:
idx2 = (h[i]/ vv[1] ).astype( np.uint32 )
val.append( vv[0][idx2[:,0],idx2[:,1],idx2[:,2]] )
except IndexError, e:
print "Index error occured when trying extract scalar values for measure", k
print "--> Discard fiber with index", i, "Exception: ", e
print "----"
da = np.concatenate( val )
di[k + '_mean'] = float(da.mean())
di[k + '_std'] = float(da.std())
del da
del val
G.add_edge(u,v, di)
# storing network
if 'gPickle' in output_types:
nx.write_gpickle(G, 'connectome_%s.gpickle' % parkey)
if 'mat' in output_types:
# edges
size_edges = (parval['number_of_regions'],parval['number_of_regions'])
edge_keys = G.edges(data=True)[0][2].keys()
edge_struct = {}
for edge_key in edge_keys:
edge_struct[edge_key] = nx.to_numpy_matrix(G,weight=edge_key)
# nodes
size_nodes = parval['number_of_regions']
node_keys = G.nodes(data=True)[0][1].keys()
node_struct = {}
for node_key in node_keys:
if node_key == 'dn_position':
node_arr = np.zeros([size_nodes,3],dtype=np.float)
else:
node_arr = np.zeros(size_nodes,dtype=np.object_)
node_n = 0
for _,node_data in G.nodes(data=True):
node_arr[node_n] = node_data[node_key]
node_n += 1
node_struct[node_key] = node_arr
scipy.io.savemat('connectome_%s.mat' % parkey, mdict={'sc':edge_struct,'nodes':node_struct})
if 'graphml' in output_types:
g2 = nx.Graph()
for u_gml,v_gml,d_gml in G.edges_iter(data=True):
g2.add_edge(u_gml,v_gml,d_gml)
for u_gml,d_gml in G.nodes(data=True):
g2.add_node(u_gml,{'dn_correspondence_id':d_gml['dn_correspondence_id'],
'dn_fsname':d_gml['dn_fsname'],
'dn_hemisphere':d_gml['dn_hemisphere'],
'dn_name':d_gml['dn_name'],
'dn_position_x':float(d_gml['dn_position'][0]),
'dn_position_y':float(d_gml['dn_position'][1]),
'dn_position_z':float(d_gml['dn_position'][2]),
'dn_region':d_gml['dn_region']})
nx.write_graphml(g2,'connectome_%s.graphml' % parkey)
print("Storing final fiber length array")
fiberlabels_fname = 'final_fiberslength_%s.npy' % str(parkey)
np.save(fiberlabels_fname, final_fiberlength_array)
print("Storing all fiber labels (with orphans)")
fiberlabels_fname = 'filtered_fiberslabel_%s.npy' % str(parkey)
np.save(fiberlabels_fname, np.array(fiberlabels, dtype = np.int32), )
print("Storing final fiber labels (no orphans)")
fiberlabels_noorphans_fname = 'final_fiberlabels_%s.npy' % str(parkey)
np.save(fiberlabels_noorphans_fname, final_fiberlabels_array)
if not streamline_wrote:
print("Filtering tractography - keeping only no orphan fibers")
finalfibers_fname = 'streamline_final.trk'
save_fibers(hdr, fib, finalfibers_fname, final_fibers_idx)
# fibers[0][0,:]
# first fiber, last point, x,y,z
# fibers[0][-1,:]
def test(a):
print "me", a
#test("test")
print "number of fibers", len(fibers)
# Translate from mm to index
# endpoints[i,0,0] = int( endpoints[i,0,0] / float(voxelSize[0]))
# convert the first fiber to a sequence of voxel indices i,j,k
# idx = (fibers[0] / np.array([1,1,1]) ).astype(np.int32)
# scalar values along first fiber
# myvolumedata[idx[:,0],idx[:,1],idx[:,2]]
# compute length of first fiber
length(fibers[0])
# return indices of all the fibers from 64 to 78
idx = np.where(mylabelarray == 64.78)[0]
for i in idx:
myfiber = fiber[i]
# happy computing
# matrix F: voxels x fibers
def DC_location_correlation(Rx_Buffer):
# Detecting downchirp
SF = param_configs(1)
BW = param_configs(2)
Fs = param_configs(3)
N = int(2**SF)
num_DC = param_configs(6)
# thresholds
corr_threshold = param_configs(
8) # Threshold above which we extract all Correlation peaks
pnts_threshold = param_configs(
9) # Max. # of peaks to extract from Corrrelation Plot
DC = np.conj(sym_to_data_ang([1], N))
Downchirp_ind = []
Cross_Corr = []
st = time.perf_counter()
###################################################
# Cross Correlation with a Single downchirp
# OLD:
for i in range(length(Rx_Buffer) - length(DC) - 1):
Cross_Corr.append(np.sum(Rx_Buffer[ i : i + N ] * DC.conj()) \
/ math.sqrt(np.sum( Rx_Buffer[ i : i + N ] * Rx_Buffer[ i : i + (N) ].conj() ) *
np.sum( DC * DC.conj())))
Cross_Corr = np.array(Cross_Corr)
Cross_Corr = Cross_Corr[np.isfinite(Cross_Corr)]
corr_threshold = 4 * np.sum(np.abs(Cross_Corr)) / length(Cross_Corr)
###################################################
n_samp_array = []
peak_ind_prev = np.array([])
for i in range(math.floor(length(Cross_Corr) / N)):
# windowing Cross-Correlation (window length N samples)
wind = np.abs(Cross_Corr[i * N:(i + 1) * N])
# Extract Multiple Correlation Peaks
peak_ind_curr = get_max(wind, corr_threshold, pnts_threshold)
if (length(peak_ind_prev) != 0 and length(peak_ind_curr) != 0):
for j in range(length(peak_ind_curr)):
for k in range(length(peak_ind_prev)):
# check if combination of any two peaks in consecutive window are N samples apart
if (peak_ind_curr[j] == peak_ind_prev[k]):
n_samp_array += [
peak_ind_prev[k] + ((i - 1) * N) + 1,
peak_ind_curr[j] + ((i) * N) + 1
]
# This extracts a list of all peaks that are N samples apart
peak_ind_prev = peak_ind_curr
n_samp_array = np.array(n_samp_array)
for i in range(length(n_samp_array)):
c = 0
ind_arr = [n_samp_array[i], n_samp_array[i] + N]
for j in range(len(ind_arr)):
c = c + np.sum(n_samp_array == ind_arr[j])
# Find from the list all the peaks that appear consecutively for
# more than 2 windows (Downchirp should give 3 peaks, N sampled apart)
if (c >= 2):
Downchirp_ind += [ind_arr]
# filter Downchirps that are with in 3 samples (same pkt detected twice due to peak energy spread)
# filter Downchirps that are with in 3 samples (same pkt detected twice due to peak energy spread)
temp = []
indices = [np.zeros(math.floor(num_DC))] + Downchirp_ind
indices = np.array(indices)
for i in range(1, indices.shape[0]):
if (len(temp) == 0):
temp.append(indices[i])
else:
if (np.min(
np.abs(indices[np.unravel_index(i, indices.shape, 'F')] -
np.array(temp)[:, 0])) > 3):
temp.append(indices[i])
Downchirp_ind = np.array(temp)
return Downchirp_ind
def UC_location_corr_DC_based(Data,DC_ind):
# chirp variables
SF = param_configs(1)
BW = param_configs(2)
Fs = param_configs(3)
N = 2**SF
upsampling_factor = Fs/BW
# LORA pkt variables
num_preamble = param_configs(4)
num_sync = param_configs(5)
num_DC = param_configs(6)
num_data_sym = param_configs(7)
# thresholds
corr_threshold = param_configs(10)
pnts_threshold = param_configs(11)
DC = np.conj(sym_to_data_ang([1],N))
####################################
if(len(DC_ind) == 0):
return
## Find list of Potential Preambles from list of Downchirps Detected
pot_pream_ind = []
for i in range(len(DC_ind)):
if DC_ind[i,0] - ((num_preamble+num_sync)*N) < 1:
continue
pot_pream_ind.append(np.arange(DC_ind[i,0] - ((num_preamble + num_sync)*N), DC_ind[i,0]- ((num_sync)*N) + 1, N))
pot_pream_ind = np.array(pot_pream_ind)
Upchirp_ind = []
# Cross Correlation with a Single UpChirp
temp_wind = []
for j in range(pot_pream_ind.shape[0]):
if pot_pream_ind[j,0] - N <= 0:
continue
Data_buffer = []
Data_buffer = Data[int(pot_pream_ind[j,0] - N) : int(pot_pream_ind[j,-1] + N)]
temp = [0+0j]
for i in range(length(Data_buffer) - length(DC)):
temp.append(np.sum(np.multiply(Data_buffer[i + 1 : i + N + 1], DC[:N])) \
/ math.sqrt(np.sum(Data_buffer[i + 1: i + N + 1] * Data_buffer[i + 1 : i + N + 1].conj() ) * \
np.sum( DC[:N] * DC[:N].conj())))
temp_wind.append(temp)
temp_wind = np.array(temp_wind)
array_stack = []
# iterate over each Downchirp Detected
for m in range(temp_wind.shape[0]):
n_samp_array = []
peak_ind_prev = np.array([])
for i in range(math.floor(length(temp_wind)/N)):
# windowing Cross-Correlation Arrays correponsing to each pkt (window length N samples)
wind = abs(temp_wind[m,i*N + 1 : (i+1) * N])
peak_ind_curr = get_max(wind,corr_threshold,pnts_threshold)
if length(peak_ind_prev) != 0 and length(peak_ind_curr) != 0:
for j in range(length(peak_ind_curr)):
for k in range(length(peak_ind_prev)):
# check if combination of any two peaks in consecutive window are N samples apart
if abs(peak_ind_curr[j]) == abs(peak_ind_prev[k]):
n_samp_array.append(peak_ind_prev[k]+((i-1)*N)+(pot_pream_ind[m,0]-N-1) + 3) # add 3 to account for zero indexing
# This extracts a list of all peaks that are N samples apart
peak_ind_prev = peak_ind_curr
array_stack.append(n_samp_array)
array_stack = np.array(array_stack)
for m in range(len(array_stack)):
n_samp_array = np.array(array_stack[m])
for i in range(length(n_samp_array)):
c = 0
ind_arr = np.arange(n_samp_array[i] + N, n_samp_array[i] + N + (num_preamble-2)*N + 1, N)
for j in range(len(ind_arr)):
c = c + np.sum( n_samp_array == ind_arr[j] )
# Find from the list all the peaks that appear consecutively for
# more than 6 windows (Upchirp should give 8 peaks, N sampled apart)
if c >= 6:
if len(Upchirp_ind) != 0:
if np.sum(np.array(Upchirp_ind)[:,0] == n_samp_array[i]) != 1:
Upchirp_ind.append(np.concatenate([[n_samp_array[i]],ind_arr]))
else:
Upchirp_ind.append(np.concatenate([[n_samp_array[i]], ind_arr]))
# filter Upchirps that are with in 5 samples (same pkt detected multiple times due to peak energy spread)
temp = []
Upchirp_ind = np.array(Upchirp_ind)
indices = np.concatenate([np.zeros((1,num_preamble)), Upchirp_ind])
for i in range(1, indices.shape[0]):
if len(temp) == 0:
temp.append(indices[i,:])
else:
if min(abs(indices[i][0] - np.array(temp)[:,0])) > 5:
temp.append(indices[i,:])
temp = np.array(temp)
Upchirp_ind = temp
return [Upchirp_ind]
