diff options
author | Nick Foster <nick@ettus.com> | 2011-06-01 09:09:24 -0700 |
---|---|---|
committer | Nick Foster <nick@ettus.com> | 2011-06-01 09:09:24 -0700 |
commit | e081fe1f098c09a2661e954bec3e511dfc28c897 (patch) | |
tree | a085cf20aeb9975609c30fd638cc05e30713dd3f /host | |
parent | c41328816b96fdf01119de4960069ebbc201da8d (diff) | |
download | uhd-e081fe1f098c09a2661e954bec3e511dfc28c897.tar.gz uhd-e081fe1f098c09a2661e954bec3e511dfc28c897.tar.bz2 uhd-e081fe1f098c09a2661e954bec3e511dfc28c897.zip |
updated sync doc for clarity & examples
Diffstat (limited to 'host')
-rw-r--r-- | host/docs/sync.rst | 106 |
1 files changed, 64 insertions, 42 deletions
diff --git a/host/docs/sync.rst b/host/docs/sync.rst index 15f3b10cf..7e9ea11f4 100644 --- a/host/docs/sync.rst +++ b/host/docs/sync.rst @@ -5,18 +5,19 @@ UHD - Synchronization Application Notes .. contents:: Table of Contents The following application notes explain how to synchronize multiple USRPs -with the goal of transmitting or receiving time-aligned samples. +with the goal of transmitting or receiving time-aligned samples for MIMO +or other applications requiring multiple USRPs operating synchronously. -**Note:** The following synchronization notes do not apply to USRP1 +**Note:** The following synchronization notes do not apply to USRP1, which does not support the advanced features available in newer products. ------------------------------------------------------------------------ Common reference signals ------------------------------------------------------------------------ -USRPs take two reference signals in order to synchronize clocks and time. +USRPs take two reference signals in order to synchronize clocks and time: -* A 10MHz reference to synchronize the clocks across devices. -* A pulse-per-second to synchronize the time across devices. +* A 10MHz reference to provide a single frequency reference for both devices, and +* A pulse-per-second (1PPS) to synchronize the sample time across devices. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Provide reference signals @@ -24,14 +25,16 @@ Provide reference signals Most USRPs have front panel SMA connectors to source these reference signals. Typically, these signals are provided by an external GPSDO. Some USRP models can provide these signals from an optional internal GPSDO. -For user's generating their own signals, +For users generating their own signals, the pulse-per-second should be clocked from the 10MHz reference. See the application notes for your device for specific signal requirements. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Set the clock configuration ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -The user should set the "external" clock configuration when configuring the USRP device object. +In order to synchronize to an external clock, configure the USRP device +using the "external" clock configuration: + :: usrp->set_clock_config(uhd::clock_config_t::external()); @@ -40,18 +43,24 @@ The user should set the "external" clock configuration when configuring the USRP Synchronizing the device time ------------------------------------------------------------------------ The purpose of the PPS signal is to synchronously latch a time into the device. -The device time may be absolute (GPS, UTC) or relative. -The user should decide what is most useful to the application. -To synchronously set the device time, -the user should determine the PPS edge, -and then make the API call set_time_next_pps(...). -The user can determine the PPS edge in a variety of ways... +You can use the set_time_next_pps(...) function to either initialize the sample time to 0, +or to an absolute time such as GPS time or UTC time. +For the purposes of synchronizing devices, +it doesn't matter what time you initialize to when using set_time_next_pps(...). + +Some GPSDOs synchronize their PPS with the rising edge of the 10MHz clock, +and some synchronize with the falling edge of the clock. +You should find out which edge your GPSDO uses, +and then pass that in as an argument to set_time_next_pps(...). + +Here are two examples of how to set the PPS time to synchronize USRPs +to a clock source. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Method 1 - poll the USRP time registers ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -One way to catch the PPS edge is to poll the "last PPS" time from the USRP device. -When the last PPS time increments, the user can determine that a PPS has occurred. +One way to initialize the PPS edge is to poll the "last PPS" time from the USRP device. +When the last PPS time increments, the user can determine that a PPS has occurred: :: @@ -66,7 +75,7 @@ Method 2 - query the GPSDO for seconds ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Most GPSDO can be configured to output a NMEA string over the serial port once every PPS. The user can wait for this string to determine the PPS edge, -and the user can also parse this string to determine GPS time. +and the user can also parse this string to determine GPS time: :: @@ -104,35 +113,48 @@ Synchronizing channel phase ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Align CORDICs in the DSP ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -The frequency translation between baseband and IF can introduce a phase ambiguity between channels. -This is due to the property that the cordic in the DSP chain can have a random phase offset. -Fortunately, the CORDIC is reset on each start-of-burst. -Therefore, the method to phase-align the cordics is to coordinate streaming at a particular time. +In order to achieve phase alignment between USRPs, the CORDICS in both +devices must be aligned with respect to each other. This is easily achieved +by issuing stream commands with a time spec property, which instructs the +streaming to begin at a specified time. Since the devices are already +synchronized via the 10MHz and PPS inputs, the streaming will start at exactly +the same time on both devices. The CORDICs are reset at each start-of-burst +command, so users should ensure that every start-of-burst also has a time spec set. + +For receive, a burst is started when the user issues a stream command. This stream command should have a time spec set. + +:: + + uhd::stream_cmd_t stream_cmd(uhd::stream_cmd_t::STREAM_MODE_NUM_SAMPS_AND_DONE); + stream_cmd.num_samps = spb; + stream_cmd.stream_now = false; + stream_cmd.time_spec = uhd::time_spec_t(seconds_in_future); + usrp->issue_stream_cmd(stream_cmd); + +For transmit, a burst is started when the user calls send(). The metadata should have a time spec and start of burst set. + +:: + + uhd::tx_metadata_t md; + md.start_of_burst = true; + md.end_of_burst = false; + md.has_time_spec = true; + md.time_spec = uhd::time_spec_t(time_to_send); -* For receive, a burst is started when the user issues a stream command. This stream command should have a time spec set. -* For transmit, a burst is started when the user calls send. The metadata should have a time spec and start of burst set. + //send a single packet + size_t num_tx_samps = usrp->get_device()->send( + buffs, samps_to_send, md, + uhd::io_type_t::COMPLEX_FLOAT32, + uhd::device::SEND_MODE_ONE_PACKET, timeout + ); ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Align LOs in the front-end ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ After tuning the RF front-ends, -each local oscillator will have a random phase offset. -Some daughterboards have mechanisms to phase-align the oscillators, -however, these mechanisms are not currently implemented. -In any case, LO phase-alignment will not forgo the need for calibration. - -**RX Calibration** - -1) Tune the setup. -2) Receive known signals. -3) Capture calibration data. -4) Do not re-tune. -5) Receive live data. - -**TX Calibration** - -1) Tune the setup. -2) Transmit known signals. -3) Capture calibration data. -4) Do not re-tune. -5) Transmit live data. +each local oscillator may have a random phase offset due to the dividers +in the VCO/PLL chains. This offset will remain constant after the device +has been initialized, and will remain constant until the device is closed +or re-tuned. This phase offset is typically removed by the user in MIMO +applications, using a training sequence to estimate the offset. It will +be necessary to re-align the LOs after each tune command. |