dsp rework: tx trailer, scaling work (peak)

11 files changed, 110 insertions, 31 deletions

diff --git a/host/include/uhd/stream.hpp b/host/include/uhd/stream.hpp
index 0c9282545..cec2eee79 100644
--- a/host/include/uhd/stream.hpp
+++ b/host/include/uhd/stream.hpp
@@ -79,10 +79,11 @@ struct UHD_API stream_args_t{
      * The args parameter is used to pass arbitrary key/value pairs.
      * Possible keys used by args (depends on implementation):
      *
-     * - scalar: an integer scaling factor used with the sc8 wire format.
-     * Use scalar=1 to harvest the lower 8-bits.
-     * Use scalar=256 to harvest the upper 8-bits.
-     * Any scalar in-between is also a possibility.
+     * - peak: specifies a fractional sample level to calculate scaling with the sc8 wire format.
+     * When using sc8 samples over the wire, the device must scale samples
+     * (both on the host and in the device) to satisfy the dynamic range needs.
+     * The peak value specifies a fraction of the maximum sample level (1.0 = 100%).
+     * Set peak to max_sample_level/full_scale_level to ensure optimum dynamic range.
      *
      * - underflow_policy: how the TX DSP should recover from underflow.
      * Possible options are "next_burst" or "next_packet".
diff --git a/host/lib/convert/convert_common.hpp b/host/lib/convert/convert_common.hpp
index 699d6301b..aab59de4b 100644
--- a/host/lib/convert/convert_common.hpp
+++ b/host/lib/convert/convert_common.hpp
@@ -190,4 +190,52 @@ static UHD_INLINE void item32_sc8_to_fc64(item32_t item, fc64_t &out0, fc64_t &o
     );
 }
 
+/***********************************************************************
+ * Convert complex char to items32 sc8 buffer
+ **********************************************************************/
+static UHD_INLINE item32_t sc8_to_item32_sc8(sc8_t in0, sc8_t in1, double){
+    return
+        (item32_t(boost::uint8_t(in0.real())) << 8) |
+        (item32_t(boost::uint8_t(in0.imag())) << 0) |
+        (item32_t(boost::uint8_t(in1.real())) << 24) |
+        (item32_t(boost::uint8_t(in1.imag())) << 16)
+    ;
+}
+
+/***********************************************************************
+ * Convert complex short to items32 sc8 buffer
+ **********************************************************************/
+static UHD_INLINE item32_t sc16_to_item32_sc8(sc16_t in0, sc16_t in1, double){
+    return
+        (item32_t(boost::uint8_t(in0.real())) << 8) |
+        (item32_t(boost::uint8_t(in0.imag())) << 0) |
+        (item32_t(boost::uint8_t(in1.real())) << 24) |
+        (item32_t(boost::uint8_t(in1.imag())) << 16)
+    ;
+}
+
+/***********************************************************************
+ * Convert complex float to items32 sc8 buffer
+ **********************************************************************/
+static UHD_INLINE item32_t fc32_to_item32_sc8(fc32_t in0, fc32_t in1, double scale_factor){
+    return
+        (item32_t(boost::uint8_t(in0.real()*float(scale_factor))) << 8) |
+        (item32_t(boost::uint8_t(in0.imag()*float(scale_factor))) << 0) |
+        (item32_t(boost::uint8_t(in1.real()*float(scale_factor))) << 24) |
+        (item32_t(boost::uint8_t(in1.imag()*float(scale_factor))) << 16)
+    ;
+}
+
+/***********************************************************************
+ * Convert complex double to items32 sc8 buffer
+ **********************************************************************/
+static UHD_INLINE item32_t fc64_to_item32_sc8(fc64_t in0, fc64_t in1, double scale_factor){
+    return
+        (item32_t(boost::uint8_t(in0.real()*(scale_factor))) << 8) |
+        (item32_t(boost::uint8_t(in0.imag()*(scale_factor))) << 0) |
+        (item32_t(boost::uint8_t(in1.real()*(scale_factor))) << 24) |
+        (item32_t(boost::uint8_t(in1.imag()*(scale_factor))) << 16)
+    ;
+}
+
 #endif /* INCLUDED_LIBUHD_CONVERT_COMMON_HPP */
diff --git a/host/lib/convert/convert_with_tables.cpp b/host/lib/convert/convert_with_tables.cpp
index c033a2959..2379739a7 100644
--- a/host/lib/convert/convert_with_tables.cpp
+++ b/host/lib/convert/convert_with_tables.cpp
@@ -68,24 +68,19 @@ class convert_sc8_item32_1_to_fcxx_1 : public converter{
 public:
     convert_sc8_item32_1_to_fcxx_1(void): _table(sc16_table_len){}
 
-    void set_scalar(const double scalar){
-
-        //special case, this is converts sc8 to sc16,
-        //use the scale-factor but no normalization
+    //special case for sc16 type, 32767 undoes float normalization
+    static type conv(const boost::int8_t &num, const double scalar){
         if (sizeof(type) == sizeof(s16_t)){
-            for (size_t i = 0; i < sc16_table_len; i++){
-                const boost::uint16_t val = tohost(boost::uint16_t(i & 0xffff));
-                const type real = type(boost::math::iround(boost::int8_t(val >> 8)*scalar*32767));
-                const type imag = type(boost::math::iround(boost::int8_t(val >> 0)*scalar*32767));
-                _table[i] = std::complex<type>(real, imag);
-            }
-            return;
+            return type(boost::math::iround(num*scalar*32767));
         }
+        return type(num*scalar);
+    }
 
+    void set_scalar(const double scalar){
         for (size_t i = 0; i < sc16_table_len; i++){
             const boost::uint16_t val = tohost(boost::uint16_t(i & 0xffff));
-            const type real = type(boost::int8_t(val >> 8)*scalar);
-            const type imag = type(boost::int8_t(val >> 0)*scalar);
+            const type real = conv(boost::int8_t(val >> 8), scalar);
+            const type imag = conv(boost::int8_t(val >> 0), scalar);
             _table[i] = std::complex<type>(real, imag);
         }
     }
diff --git a/host/lib/convert/gen_convert_general.py b/host/lib/convert/gen_convert_general.py
index b8d64aa4b..a6ae3f810 100644
--- a/host/lib/convert/gen_convert_general.py
+++ b/host/lib/convert/gen_convert_general.py
@@ -92,6 +92,22 @@ DECLARE_CONVERTER(sc8_item32_$(end), 1, $(cpu_type), 1, PRIORITY_GENERAL){
         item32_sc8_to_$(cpu_type)(item_n, output[num_samps-1], dummy, scale_factor);
     }
 }
+
+DECLARE_CONVERTER($(cpu_type), 1, sc8_item32_$(end), 1, PRIORITY_GENERAL){
+    const $(cpu_type)_t *input = reinterpret_cast<const $(cpu_type)_t *>(inputs[0]);
+    item32_t *output = reinterpret_cast<item32_t *>(outputs[0]);
+
+    const size_t num_pairs = nsamps/2;
+    for (size_t i = 0, j = 0; i < num_pairs; i++, j+=2){
+        const item32_t item = $(cpu_type)_to_item32_sc8(input[j], input[j+1], scale_factor);
+        output[i] = $(to_wire)(item);
+    }
+
+    if (nsamps != num_pairs*2){
+        const item32_t item = $(cpu_type)_to_item32_sc8(input[nsamps-1], 0, scale_factor);
+        output[num_pairs] = $(to_wire)(item);
+    }
+}
 """
 
 TMPL_CONV_USRP1_COMPLEX = """
diff --git a/host/lib/transport/gen_vrt_if_packet.py b/host/lib/transport/gen_vrt_if_packet.py
index 245a7ddbd..245a7ddbd 100755..100644
--- a/host/lib/transport/gen_vrt_if_packet.py
+++ b/host/lib/transport/gen_vrt_if_packet.py
diff --git a/host/lib/transport/super_send_packet_handler.hpp b/host/lib/transport/super_send_packet_handler.hpp
index 5ed8e0143..57304a7d4 100644
--- a/host/lib/transport/super_send_packet_handler.hpp
+++ b/host/lib/transport/super_send_packet_handler.hpp
@@ -134,7 +134,7 @@ public:
         vrt::if_packet_info_t if_packet_info;
         if_packet_info.has_sid = false;
         if_packet_info.has_cid = false;
-        if_packet_info.has_tlr = false;
+        if_packet_info.has_tlr = true;
         if_packet_info.has_tsi = metadata.has_time_spec;
         if_packet_info.has_tsf = metadata.has_time_spec;
         if_packet_info.tsi     = boost::uint32_t(metadata.time_spec.get_full_secs());
diff --git a/host/lib/usrp/b100/io_impl.cpp b/host/lib/usrp/b100/io_impl.cpp
index 6e9e017aa..9474268ef 100644
--- a/host/lib/usrp/b100/io_impl.cpp
+++ b/host/lib/usrp/b100/io_impl.cpp
@@ -263,6 +263,7 @@ tx_streamer::sptr b100_impl::get_tx_stream(const uhd::stream_args_t &args_){
     //calculate packet size
     static const size_t hdr_size = 0
         + vrt::max_if_hdr_words32*sizeof(boost::uint32_t)
+        + sizeof(vrt::if_packet_info_t().tlr) //forced to have trailer
         - sizeof(vrt::if_packet_info_t().cid) //no class id ever used
     ;
     static const size_t bpp = 2048 - hdr_size;
diff --git a/host/lib/usrp/cores/rx_dsp_core_200.cpp b/host/lib/usrp/cores/rx_dsp_core_200.cpp
index b2bf934bc..7b35daa42 100644
--- a/host/lib/usrp/cores/rx_dsp_core_200.cpp
+++ b/host/lib/usrp/cores/rx_dsp_core_200.cpp
@@ -63,8 +63,7 @@ public:
     {
         //init to something so update method has reasonable defaults
         _scaling_adjustment = 1.0;
-        _extra_scaling = 1.0;
-        _fxpt_scalar_correction = 1.0;
+        _dsp_extra_scaling = 1.0;
 
         //This is a hack/fix for the lingering packet problem.
         //The caller should also flush the recv transports
@@ -187,14 +186,14 @@ public:
     }
 
     void update_scalar(void){
-        const double target_scalar = (1 << 16)*_scaling_adjustment/_extra_scaling;
+        const double target_scalar = (1 << 16)*_scaling_adjustment/_dsp_extra_scaling;
         const boost::int32_t actual_scalar = boost::math::iround(target_scalar);
         _fxpt_scalar_correction = target_scalar/actual_scalar; //should be small
         _iface->poke32(REG_DSP_RX_SCALE_IQ, actual_scalar);
     }
 
     double get_scaling_adjustment(void){
-        return _fxpt_scalar_correction*_extra_scaling/32767.;
+        return _fxpt_scalar_correction*_host_extra_scaling/32767.;
     }
 
     double set_freq(const double freq_){
@@ -230,11 +229,15 @@ public:
         unsigned format_word = 0;
         if (stream_args.otw_format == "sc16"){
             format_word = 0;
-            _extra_scaling = 1.0;
+            _dsp_extra_scaling = 1.0;
+            _host_extra_scaling = 1.0;
         }
         else if (stream_args.otw_format == "sc8"){
             format_word = (1 << 0);
-            _extra_scaling = stream_args.args.cast<double>("scalar", 1.0);
+            double peak = stream_args.args.cast<double>("peak", 1.0);
+            peak = std::max(peak, 1.0/256);
+            _host_extra_scaling = peak*256;
+            _dsp_extra_scaling = peak*256;
         }
         else throw uhd::value_error("USRP RX cannot handle requested wire format: " + stream_args.otw_format);
 
@@ -248,7 +251,7 @@ private:
     const size_t _dsp_base, _ctrl_base;
     double _tick_rate, _link_rate;
     bool _continuous_streaming;
-    double _scaling_adjustment, _extra_scaling, _fxpt_scalar_correction;
+    double _scaling_adjustment, _dsp_extra_scaling, _host_extra_scaling, _fxpt_scalar_correction;
     const boost::uint32_t _sid;
 };
 
diff --git a/host/lib/usrp/cores/tx_dsp_core_200.cpp b/host/lib/usrp/cores/tx_dsp_core_200.cpp
index 2c5b86bfc..15358e184 100644
--- a/host/lib/usrp/cores/tx_dsp_core_200.cpp
+++ b/host/lib/usrp/cores/tx_dsp_core_200.cpp
@@ -18,6 +18,7 @@
 #include "tx_dsp_core_200.hpp"
 #include <uhd/types/dict.hpp>
 #include <uhd/exception.hpp>
+#include <uhd/utils/msg.hpp>
 #include <uhd/utils/algorithm.hpp>
 #include <boost/assign/list_of.hpp>
 #include <boost/math/special_functions/round.hpp>
@@ -35,6 +36,7 @@
 #define REG_TX_CTRL_POLICY          _ctrl_base + 12
 #define REG_TX_CTRL_CYCLES_PER_UP   _ctrl_base + 16
 #define REG_TX_CTRL_PACKETS_PER_UP  _ctrl_base + 20
+#define REG_TX_CTRL_FORMAT          _ctrl_base + 24
 
 #define FLAG_TX_CTRL_POLICY_WAIT          (0x1 << 0)
 #define FLAG_TX_CTRL_POLICY_NEXT_PACKET   (0x1 << 1)
@@ -60,8 +62,7 @@ public:
     {
         //init to something so update method has reasonable defaults
         _scaling_adjustment = 1.0;
-        _extra_scaling = 1.0;
-        _fxpt_scalar_correction = 1.0;
+        _dsp_extra_scaling = 1.0;
 
         //init the tx control registers
         this->clear();
@@ -134,14 +135,14 @@ public:
     }
 
     void update_scalar(void){
-        const double target_scalar = (1 << 17)*_scaling_adjustment/_extra_scaling;
+        const double target_scalar = (1 << 17)*_scaling_adjustment/_dsp_extra_scaling;
         const boost::int32_t actual_scalar = boost::math::iround(target_scalar);
         _fxpt_scalar_correction = target_scalar/actual_scalar; //should be small
         _iface->poke32(REG_DSP_TX_SCALE_IQ, actual_scalar);
     }
 
     double get_scaling_adjustment(void){
-        return _fxpt_scalar_correction*_extra_scaling*32767.;
+        return _fxpt_scalar_correction*_host_extra_scaling*32767.;
     }
 
     double set_freq(const double freq_){
@@ -175,13 +176,25 @@ public:
     void setup(const uhd::stream_args_t &stream_args){
         if (not stream_args.args.has_key("noclear")) this->clear();
 
+        unsigned format_word = 0;
         if (stream_args.otw_format == "sc16"){
-            _extra_scaling = 1.0;
+            format_word = 0;
+            _dsp_extra_scaling = 1.0;
+            _host_extra_scaling = 1.0;
+        }
+        else if (stream_args.otw_format == "sc8"){
+            format_word = (1 << 0);
+            double peak = stream_args.args.cast<double>("peak", 1.0);
+            peak = std::max(peak, 1.0/256);
+            _host_extra_scaling = 1.0/peak/256;
+            _dsp_extra_scaling = 1.0/peak;
         }
         else throw uhd::value_error("USRP TX cannot handle requested wire format: " + stream_args.otw_format);
 
         this->update_scalar();
 
+        _iface->poke32(REG_TX_CTRL_FORMAT, format_word);
+
         if (stream_args.args.has_key("underflow_policy")){
             this->set_underflow_policy(stream_args.args["underflow_policy"]);
         }
@@ -191,7 +204,7 @@ private:
     wb_iface::sptr _iface;
     const size_t _dsp_base, _ctrl_base;
     double _tick_rate, _link_rate;
-    double _scaling_adjustment, _extra_scaling, _fxpt_scalar_correction;
+    double _scaling_adjustment, _dsp_extra_scaling, _host_extra_scaling, _fxpt_scalar_correction;
     const boost::uint32_t _sid;
 };
 
diff --git a/host/lib/usrp/e100/io_impl.cpp b/host/lib/usrp/e100/io_impl.cpp
index 89a4ed00b..53f783a3f 100644
--- a/host/lib/usrp/e100/io_impl.cpp
+++ b/host/lib/usrp/e100/io_impl.cpp
@@ -339,6 +339,7 @@ tx_streamer::sptr e100_impl::get_tx_stream(const uhd::stream_args_t &args_){
     //calculate packet size
     static const size_t hdr_size = 0
         + vrt::max_if_hdr_words32*sizeof(boost::uint32_t)
+        + sizeof(vrt::if_packet_info_t().tlr) //forced to have trailer
         - sizeof(vrt::if_packet_info_t().cid) //no class id ever used
     ;
     static const size_t bpp = _data_transport->get_send_frame_size() - hdr_size;
diff --git a/host/lib/usrp/usrp2/io_impl.cpp b/host/lib/usrp/usrp2/io_impl.cpp
index 58a5df2a5..6dcec6f1f 100644
--- a/host/lib/usrp/usrp2/io_impl.cpp
+++ b/host/lib/usrp/usrp2/io_impl.cpp
@@ -450,6 +450,7 @@ tx_streamer::sptr usrp2_impl::get_tx_stream(const uhd::stream_args_t &args_){
     //calculate packet size
     static const size_t hdr_size = 0
         + vrt::max_if_hdr_words32*sizeof(boost::uint32_t)
+        + sizeof(vrt::if_packet_info_t().tlr) //forced to have trailer
         + vrt_send_header_offset_words32*sizeof(boost::uint32_t)
         - sizeof(vrt::if_packet_info_t().cid) //no class id ever used
     ;