def setUp(self): self.tb = gr.top_block() self.src = gr.vector_source_b((0, 1, 2, 3, 4, 5, 6, 7, 8, 9), True, 10) self.head = gr.head(10, 2) seq = (3, 7, 5, 8, 6, 1, 9, 4, 2, 0) self.interleaver = drm.interleaver_vbvb(seq) self.snk = gr.vector_sink_b(10) self.tb.connect(self.src, self.head, self.interleaver, self.snk)
def __init__(self, vlen_in=1, vlen_out=1, n_tailbits=6, denom_mother_code_rate=6, gen_poly=(91, 121, 101, 91, 121, 101), N=1, bits_per_symbol=0, pp=0, pp_tail=0, interl_seq=range(2), map_tab=0): gr.hier_block2.__init__( self, "DRM MLC 4-QAM", gr.io_signature(1, 1, gr.sizeof_char*vlen_in), gr.io_signature(1, 1, gr.sizeof_gr_complex*vlen_out), ) ################################################## # Parameters ################################################## self.vlen_in = vlen_in self.vlen_out = vlen_out self.n_tailbits = n_tailbits self.denom_mother_code_rate = denom_mother_code_rate self.gen_poly = gen_poly self.N = N self.bits_per_symbol = bits_per_symbol self.pp = pp self.pp_tail = pp_tail self.interl_seq = interl_seq self.map_tab = map_tab ################################################## # Blocks ################################################## self.trellis_encoder_xx_0 = trellis.encoder_bb(trellis.fsm(1, denom_mother_code_rate, gen_poly), 0) self.gr_vector_to_stream_0 = gr.vector_to_stream(gr.sizeof_char*1, vlen_in + n_tailbits) self.gr_unpack_k_bits_bb_0 = gr.unpack_k_bits_bb(denom_mother_code_rate) self.gr_stream_to_vector_0 = gr.stream_to_vector(gr.sizeof_char*1, (vlen_in + n_tailbits) * denom_mother_code_rate) self.drm_qam_map_vbvc_0 = drm.qam_map_vbvc(map_tab, bits_per_symbol, vlen_out, 1) self.drm_punct_vbvb_0 = drm.punct_vbvb(pp, pp_tail, (vlen_in + n_tailbits) * denom_mother_code_rate, vlen_out * 2, n_tailbits * denom_mother_code_rate) self.drm_interleaver_vbvb_0 = drm.interleaver_vbvb((interl_seq)) self.add_tailbits_vbvb_0 = drm.add_tailbits_vbvb(vlen_in, n_tailbits) ################################################## # Connections ################################################## self.connect((self, 0), (self.add_tailbits_vbvb_0, 0)) self.connect((self.add_tailbits_vbvb_0, 0), (self.gr_vector_to_stream_0, 0)) self.connect((self.gr_vector_to_stream_0, 0), (self.trellis_encoder_xx_0, 0)) self.connect((self.trellis_encoder_xx_0, 0), (self.gr_unpack_k_bits_bb_0, 0)) self.connect((self.gr_unpack_k_bits_bb_0, 0), (self.gr_stream_to_vector_0, 0)) self.connect((self.gr_stream_to_vector_0, 0), (self.drm_punct_vbvb_0, 0)) self.connect((self.drm_punct_vbvb_0, 0), (self.drm_interleaver_vbvb_0, 0)) self.connect((self.drm_interleaver_vbvb_0, 0), (self.drm_qam_map_vbvc_0, 0)) self.connect((self.drm_qam_map_vbvc_0, 0), (self, 0))
def __init__(self, vlen_in=1, vlen_out=1, n_tailbits=6, denom_mother_code_rate=6, gen_poly=(91, 121, 101, 91, 121, 101), bits_per_symbol=0, N=1, map_tab=0, pp_0=0, pp_0_tail=0, pp_1=0, pp_1_tail=0, part_len_top=1, part_len_bot=1, M_total=0, interl_seq_0_2=range(2), interl_seq_1_2=range(2)): gr.hier_block2.__init__( self, "DRM MLC 16-QAM", gr.io_signature(1, 1, gr.sizeof_char*vlen_in), gr.io_signature(1, 1, gr.sizeof_gr_complex*vlen_out), ) ################################################## # Parameters ################################################## self.vlen_in = vlen_in self.vlen_out = vlen_out self.n_tailbits = n_tailbits self.denom_mother_code_rate = denom_mother_code_rate self.gen_poly = gen_poly self.bits_per_symbol = bits_per_symbol self.N = N self.map_tab = map_tab self.pp_0 = pp_0 self.pp_0_tail = pp_0_tail self.pp_1 = pp_1 self.pp_1_tail = pp_1_tail self.part_len_top = part_len_top self.part_len_bot = part_len_bot self.M_total = M_total self.interl_seq_0_2 = interl_seq_0_2 self.interl_seq_1_2 = interl_seq_1_2 ################################################## # Blocks ################################################## self.trellis_encoder_xx_0_0 = trellis.encoder_bb(trellis.fsm(1, denom_mother_code_rate, gen_poly), 0) self.trellis_encoder_xx_0 = trellis.encoder_bb(trellis.fsm(1, denom_mother_code_rate, gen_poly), 0) self.gr_vector_to_stream_1_0 = gr.vector_to_stream(gr.sizeof_char*1, part_len_bot+ n_tailbits) self.gr_vector_to_stream_1 = gr.vector_to_stream(gr.sizeof_char*1, part_len_top + n_tailbits) self.gr_unpack_k_bits_bb_0_0 = gr.unpack_k_bits_bb(denom_mother_code_rate) self.gr_unpack_k_bits_bb_0 = gr.unpack_k_bits_bb(denom_mother_code_rate) self.gr_stream_to_vector_0_0 = gr.stream_to_vector(gr.sizeof_char*1, (part_len_bot + n_tailbits) * denom_mother_code_rate) self.gr_stream_to_vector_0 = gr.stream_to_vector(gr.sizeof_char*1, (part_len_top + n_tailbits) * denom_mother_code_rate) self.drm_qam_map_vbvc_0 = drm.qam_map_vbvc(map_tab, bits_per_symbol, vlen_out, 2) self.drm_punct_vbvb_0_0 = drm.punct_vbvb(pp_1, pp_1_tail, (part_len_bot + n_tailbits) * denom_mother_code_rate, vlen_out * 2, n_tailbits * denom_mother_code_rate) self.drm_punct_vbvb_0 = drm.punct_vbvb(pp_0, pp_0_tail, (part_len_top + n_tailbits) * denom_mother_code_rate, vlen_out * 2, n_tailbits * denom_mother_code_rate) self.drm_partitioning_16_vbvb_0 = drm.partitioning_vbvb(vlen_in, M_total) self.drm_interleaver_vbvb_0_0 = drm.interleaver_vbvb((interl_seq_1_2)) self.drm_interleaver_vbvb_0 = drm.interleaver_vbvb((interl_seq_0_2)) self.add_tailbits_vbvb_0_0 = drm.add_tailbits_vbvb(part_len_bot, n_tailbits) self.add_tailbits_vbvb_0 = drm.add_tailbits_vbvb(part_len_top, n_tailbits) ################################################## # Connections ################################################## self.connect((self.drm_interleaver_vbvb_0, 0), (self.drm_qam_map_vbvc_0, 0)) self.connect((self.drm_qam_map_vbvc_0, 0), (self, 0)) self.connect((self.trellis_encoder_xx_0, 0), (self.gr_unpack_k_bits_bb_0, 0)) self.connect((self.gr_stream_to_vector_0, 0), (self.drm_punct_vbvb_0, 0)) self.connect((self.gr_unpack_k_bits_bb_0, 0), (self.gr_stream_to_vector_0, 0)) self.connect((self.drm_punct_vbvb_0, 0), (self.drm_interleaver_vbvb_0, 0)) self.connect((self.trellis_encoder_xx_0_0, 0), (self.gr_unpack_k_bits_bb_0_0, 0)) self.connect((self.gr_stream_to_vector_0_0, 0), (self.drm_punct_vbvb_0_0, 0)) self.connect((self.gr_unpack_k_bits_bb_0_0, 0), (self.gr_stream_to_vector_0_0, 0)) self.connect((self.drm_punct_vbvb_0_0, 0), (self.drm_interleaver_vbvb_0_0, 0)) self.connect((self.drm_interleaver_vbvb_0_0, 0), (self.drm_qam_map_vbvc_0, 1)) self.connect((self.gr_vector_to_stream_1, 0), (self.trellis_encoder_xx_0, 0)) self.connect((self.gr_vector_to_stream_1_0, 0), (self.trellis_encoder_xx_0_0, 0)) self.connect((self, 0), (self.drm_partitioning_16_vbvb_0, 0)) self.connect((self.add_tailbits_vbvb_0, 0), (self.gr_vector_to_stream_1, 0)) self.connect((self.add_tailbits_vbvb_0_0, 0), (self.gr_vector_to_stream_1_0, 0)) self.connect((self.drm_partitioning_16_vbvb_0, 1), (self.add_tailbits_vbvb_0_0, 0)) self.connect((self.drm_partitioning_16_vbvb_0, 0), (self.add_tailbits_vbvb_0, 0))
def __init__(self, vlen_in=1, vlen_out=1, n_tailbits=6, denom_mother_code_rate=6, gen_poly=(91, 121, 101, 91, 121, 101), bits_per_symbol=0, N=1, map_tab=0, pp_0=0, pp_0_tail=0, pp_1=0, pp_1_tail=0, part_len_top=1, part_len_bot=1, M_total=0, interl_seq_0_2=range(2), interl_seq_1_2=range(2)): gr.hier_block2.__init__( self, "DRM MLC 16-QAM", gr.io_signature(1, 1, gr.sizeof_char * vlen_in), gr.io_signature(1, 1, gr.sizeof_gr_complex * vlen_out), ) ################################################## # Parameters ################################################## self.vlen_in = vlen_in self.vlen_out = vlen_out self.n_tailbits = n_tailbits self.denom_mother_code_rate = denom_mother_code_rate self.gen_poly = gen_poly self.bits_per_symbol = bits_per_symbol self.N = N self.map_tab = map_tab self.pp_0 = pp_0 self.pp_0_tail = pp_0_tail self.pp_1 = pp_1 self.pp_1_tail = pp_1_tail self.part_len_top = part_len_top self.part_len_bot = part_len_bot self.M_total = M_total self.interl_seq_0_2 = interl_seq_0_2 self.interl_seq_1_2 = interl_seq_1_2 ################################################## # Blocks ################################################## self.trellis_encoder_xx_0_0 = trellis.encoder_bb( trellis.fsm(1, denom_mother_code_rate, gen_poly), 0) self.trellis_encoder_xx_0 = trellis.encoder_bb( trellis.fsm(1, denom_mother_code_rate, gen_poly), 0) self.gr_vector_to_stream_1_0 = gr.vector_to_stream( gr.sizeof_char * 1, part_len_bot + n_tailbits) self.gr_vector_to_stream_1 = gr.vector_to_stream( gr.sizeof_char * 1, part_len_top + n_tailbits) self.gr_unpack_k_bits_bb_0_0 = gr.unpack_k_bits_bb( denom_mother_code_rate) self.gr_unpack_k_bits_bb_0 = gr.unpack_k_bits_bb( denom_mother_code_rate) self.gr_stream_to_vector_0_0 = gr.stream_to_vector( gr.sizeof_char * 1, (part_len_bot + n_tailbits) * denom_mother_code_rate) self.gr_stream_to_vector_0 = gr.stream_to_vector( gr.sizeof_char * 1, (part_len_top + n_tailbits) * denom_mother_code_rate) self.drm_qam_map_vbvb_0 = drm.qam_map_vbvb(map_tab, bits_per_symbol, vlen_out, 2) self.drm_punct_vbvb_0_0 = drm.punct_vbvb( pp_1, pp_1_tail, (part_len_bot + n_tailbits) * denom_mother_code_rate, vlen_out * 2, n_tailbits * denom_mother_code_rate) self.drm_punct_vbvb_0 = drm.punct_vbvb( pp_0, pp_0_tail, (part_len_top + n_tailbits) * denom_mother_code_rate, vlen_out * 2, n_tailbits * denom_mother_code_rate) self.drm_partitioning_16_vbvb_0 = drm.partitioning_vbvb( vlen_in, M_total) self.drm_interleaver_vbvb_0_0 = drm.interleaver_vbvb((interl_seq_1_2)) self.drm_interleaver_vbvb_0 = drm.interleaver_vbvb((interl_seq_0_2)) self.add_tailbits_vbvb_0_0 = drm.add_tailbits_vbvb( part_len_bot, n_tailbits) self.add_tailbits_vbvb_0 = drm.add_tailbits_vbvb( part_len_top, n_tailbits) ################################################## # Connections ################################################## self.connect((self.drm_interleaver_vbvb_0, 0), (self.drm_qam_map_vbvb_0, 0)) self.connect((self.drm_qam_map_vbvb_0, 0), (self, 0)) self.connect((self.trellis_encoder_xx_0, 0), (self.gr_unpack_k_bits_bb_0, 0)) self.connect((self.gr_stream_to_vector_0, 0), (self.drm_punct_vbvb_0, 0)) self.connect((self.gr_unpack_k_bits_bb_0, 0), (self.gr_stream_to_vector_0, 0)) self.connect((self.drm_punct_vbvb_0, 0), (self.drm_interleaver_vbvb_0, 0)) self.connect((self.trellis_encoder_xx_0_0, 0), (self.gr_unpack_k_bits_bb_0_0, 0)) self.connect((self.gr_stream_to_vector_0_0, 0), (self.drm_punct_vbvb_0_0, 0)) self.connect((self.gr_unpack_k_bits_bb_0_0, 0), (self.gr_stream_to_vector_0_0, 0)) self.connect((self.drm_punct_vbvb_0_0, 0), (self.drm_interleaver_vbvb_0_0, 0)) self.connect((self.drm_interleaver_vbvb_0_0, 0), (self.drm_qam_map_vbvb_0, 1)) self.connect((self.gr_vector_to_stream_1, 0), (self.trellis_encoder_xx_0, 0)) self.connect((self.gr_vector_to_stream_1_0, 0), (self.trellis_encoder_xx_0_0, 0)) self.connect((self, 0), (self.drm_partitioning_16_vbvb_0, 0)) self.connect((self.add_tailbits_vbvb_0, 0), (self.gr_vector_to_stream_1, 0)) self.connect((self.add_tailbits_vbvb_0_0, 0), (self.gr_vector_to_stream_1_0, 0)) self.connect((self.drm_partitioning_16_vbvb_0, 1), (self.add_tailbits_vbvb_0_0, 0)) self.connect((self.drm_partitioning_16_vbvb_0, 0), (self.add_tailbits_vbvb_0, 0))