PIXIE-4 User’s Manual
V2.69
©
XIA
2015. All rights reserved.
xlix
by the DSP and for readout to the host PC, arbitrated by an FPGA. While the host is reading
the memory, spectrum increments are queued in a buffer (2K long). At maximum count rate, it
will take the DSP at least (2K * processing time) to fill the buffer and correspondingly longer
at lower count rates, while the host readout typically takes ~30 ms. Thus host readout dead
time is usually not an issue in MCA runs unless spectra are read very frequently.
In list mode runs, the Pixie-4 memory fills up after a certain number of events are acquired.
Acquisition is stopped until the memory is read out. Depending on the “buffer per spill” setting,
this is organized in one of three ways:
a)
In single buffer mode, acquisition stops after the DSP’s 8K buffer is filled and data is
read out in a slow transfer mode.
b)
In 32 buffer mode, the 8K buffer is automatically transferred to external memory. The
transfer incurs about 300µs dead time (this may be improved in the future). After 32
transfers, the external memory is full and acquisition is stopped. The external memory
is read out in a fast block transfer that takes about 30 ms
c)
In 16/16 buffer mode (or double buffer mode), the 8K buffer is also automatically
transferred to external memory, but only 16 times. Then a flag is raised so that the host
can read out the memory in a fast block transfer while acquisition continues and new
data is stored in the second half of the memory. However, since again the memory has
only a single port, host readout can not happen at the same time as the DSP transfer.
Normally, the readout will be finished before the DSP 8K buffer is filled, but at high
count rate or when recording waveforms, the transfer may have to wait until the host
finished reading the external memory, which means the acquisition is stopped and there
will be dead time in addition to the ~300 us transfer time.
Case b) is pictured in Fig. 6.11; the difference for case c) is shown with the dashed lines. The
Pixie-4 acquires several buffers of data (T1-T2, T3-T4) until the external memory is full and
read out by the host at T5. After readout, a second spill resumes at T6. The live time counter is
active while events can be acquired, which excludes memory transfer and readout. The run
time starts a bit earlier and lasts a bit longer than the live time for each buffer, and does not
switch off for out-of range or other conditions. The total time counts all time from T0 to the
end of the last spill. The Pixie 4 can not count the time between the user clicking the run start
button and T0. In case c), the acquisition resumes already at T5 while the memory is read out
independently. In a) and b), there is thus dead time in the Pixie-4 while it is waiting for the host
to read out the data. In b) and c) there is dead time while the data is transferred to external
memory. In all cases, the run is considered stopped in the Pixie-4, as opposed to the dead times
described in section 6.6.1.2. The readout dead times are omitted from the Pixie-4 live time and
run time counters (counters are stopped), but included in the “total time” counter.
Examples of host readout dead times are shown in Table 6.3. Rows A and B are copied from
Table 6.2 for comparison. When no traces are recorded, the fraction of time lost to readout is
less than 1% even up to moderately high count rates; and well below the fraction of events lost
due to Td (dominated by the filter time). When traces are recorded, the fraction of time lost
increases, but is still below the fraction of events lost due to Td.
Table 6.3. Dead times from host readout in 32 buffer/spill mode. (One channel active per
event)