AD9726
Rev. B | Page 21 of 24
Linearity CALDACs operate inversely from their input; that is,
as their binary input value increases, the magnitude of their
current contribution seen at the AD9726 output decreases. Gain
CALDACs are an exception to this. Their contribution seen at
the AD9726 output is in direct proportion to their binary input.
Gain CALDACs are also half strength as compared to linearity
CALDACs, but they are intended to be used together as a unit
and thus, together, provide twice the current adjustment range.
Calibration Memory
During production testing, the linearity of the AD9726 is
measured and optimized. Values for all CALDACs are perma-
nently stored in nonvolatile factory memory (FMEM). At
reset, all factory memory contents are transferred to static
memory. CALMEM, Bits[5:4] in Register 0x0E, indicates a
factory calibrated state (CALMEM = 10b).
It is also possible at any time to transfer the contents of FMEM
to SMEM by asserting the MEMXFER bit in Register 0x0F. The
XFERSTAT indicator bit (Bit 5 in Register 0x0F) then reports
the successful completion of the transfer cycle, and MEMXFER
is cleared.
Note that the MEMXFER bit (and SELFCAL, Bit 6, Register
0x0F) cannot be asserted if any other memory access function is
currently enabled (that is, if any one of Bits[3:0] in Register
0x0F is high). Attempting to assert MEMXFER (or SELFCAL)
in this case clears any asserted bits in Register 0x0F, but the
requested cycle does not commence.
The factory-to-static memory data transfer cycle requires a
number of DAC clock cycles. The total depends on the value of
CALCLK. This value sets a divider used to create a slow version
of the DAC clock, which is intended to extend the settling time
available to the self-calibration cycle. However, this divided
clock is also used to sequence a memory transfer cycle.
The divider is set to its maximum value with CALCLK at its
default value. A memory transfer cycle requires about 15 ms at a
DAC clock frequency of 100 MHz. This time can be reduced by
50% for every increase in the value of CALCLK.
Accessing Calibration Memory
SMEM or FMEM locations can be read at any time by setting
the SMEMRD or FMEMRD bit in SPI Register 0x0F. Address
and data information can be input and/or output through SPI
Register 0x10 and SPI Register 0x11, respectively.
SMEM locations can also be written by setting the SMEMWR
bit in Register 0x0F. Register 0x10 and Register 0x11 are again
used for addresses and data. Any time after the SMEMWR bit
has been asserted, the device reports a user-calibrated state
(CALMEM = 11b) until another action changes the calibration
memory status.
To reset static memory at any time, assert the UNCAL bit in
Register 0x0F. All SMEM locations are then reset to their
default values (63). CALMEM reports an uncalibrated state
(CALMEM = 00b). Note that UNCAL remains asserted (and
the contents of SMEM remains at default values) indefinitely.
UNCAL does not clear itself (like SWRESET) and must be
cleared by the user.
Note also that although SPI registers do not depend on the DAC
clock (they use SCLK to sequence the controller state machine),
SMEM and/or FMEM access does require a valid DAC clock.
SMEM/FMEM Read/Write Procedures
Static and factory memory is accessed through the SPI, but it is
not part of the SPI logic. For this reason, memory access requires
a valid DAC clock, while SPI register access does not.
Because the AD9726 SPI is so flexible, allowing single and
multiple byte reads and writes as well as MSB or LSB justified
data, there are a number of ways in which a user can access one
or more SMEM or FMEM locations.
To avoid potential errors, the following procedures for accessing
static or factory memory should be followed. These procedures
use only single-byte SPI commands to ensure the enabling of
addresses and the sequencing of memory access.
To read from SMEM or FMEM,
1. Ensure that Bits [3:0] of Register 0x0F are clear.
2. Begin the sequence by writing the memory address value
to Register 0x10 with a single-byte SPI write command.
3. Assert the SMEMRD or FMEMRD bit in Register 0x0F
with another single-byte SPI write command.
4. Import the contents of Register 0x11 using a single-byte
SPI read command.
5. Clear the SMEMRD or FMEMRD bit with another single-
byte command.
To wr ite to SM E M,
1. Ensure that Bits [3:0] of Register 0x0F are clear.
2. Begin the sequence by writing the data value to
Register 0x11 using a single-byte SPI write command.
3. Assert the SMEMWR bit using a single-byte SPI write
command.
4. Place the memory address value in Register 0x10 using a
single-byte SPI write command.
5. Clear the SMEMWR bit with a fourth single-byte SPI write
command.
Self-Calibration
The AD9726 features an internal self-calibration engine to
linearize the transfer function automatically. This can be very
useful at temperature extremes where factory calibration no
longer applies. The automated cycle can be initiated by asserting
the SELFCAL bit.
The self-calibration process calibrates all linearity and gain
CALDACs based upon a fixed internal reference current. Values
for all CALDACs are stored in volatile static memory. The
CALSTAT bit indicates the successful completion of the cycle,
AD9726
Rev. B | Page 22 of 24
and the SELFCAL bit is cleared. Following the cycle, the device
reports a self-calibrated state (CALMEM = 01b).
As with MEMXFER, successful assertion of the SELFCAL bit
(Bit 6 in Register 0x0F)requires that Bits[3:0] in Register 0x0F
be clear. If any of these bits are asserted (such that an
SMEM/FMEM read/write/clear state is enabled), the self-
calibration cycle does not begin.
The time required to self-calibrate is dependent on both the
DAC clock frequency and the value of CALCLK (Bits[5:0] in
Register 0x0E). Because self-calibration requires more time
than ordinary operation, the DAC clock is divided into a slower
version and used to step through the process. Time made
available to the self-calibration algorithm directly impacts its
ability to provide accurate results.
A maximum fixed division ratio (4096) corresponds to the
minimum default value of CALCLK (0). The division ratio can
be decreased by increasing the value of CALCLK. Each increase
in the value of CALCLK reduces the DAC clock division factor
(and, therefore, the time made available to self-calibration) by
50%. With CALCLK at its maximum value (7), the divide ratio
declines to its minimum value (32).
With CALCLK at its default value, self-calibration requires
approximately 100 ms at a DAC clock frequency of 100 MHz.
This time can be reduced to under 0.8 ms if CALCLK = 7. Time
scales relative to DAC clock frequency.
Performance Effects of Calibration
Harmonic distortion for low frequency outputs is primarily a
function of DAC linearity. Figure 12 to Figure 14 show the
harmonic distortion performance of the AD9726.
Figure 12 shows a 1 MHz full-scale output tone. The output
drives a unique low-pass and high-pass filter called a diplexer.
This type of filter presents a uniform 50  load to the DAC and
splits the output signal into low and high frequency paths. The
diplexer's low-pass output passes the 1 MHz fundamental but
attenuates higher frequencies, and the diplexer's high-pass out-
put passes higher frequencies and attenuates the 1 MHz funda-
mental. Figure 12 also shows the diplexer's low-pass output.
Here the noise floor is higher than the harmonic distortion
because with a high power input signal, attenuation is required
by the spectrum analyzer.
Figure 13 shows the diplexer's high pass output where the
attenuated input signal can be seen. The spectrum analyzer
attenuation is also reduced, which lowers the noise floor.
Harmonic products at integer multiples of the fundamental
are thus revealed. This is the response using the AD9726 in
an uncalibrated state.
Figure 14 shows a response using the AD9726 in a calibrated
state. Harmonic distortion due to the nonlinearities of the
digital-to-analog conversion are virtually eliminated.
SYNC LOGIC OPERATION AND PROGRAMMING
Recall that a fixed setup and hold timing relationship between
the data clock input and the data bus must be established and
maintained. Recall also that the data bus and the DAC clock
must be frequency locked. Because of the sync logic, however,
the phase relationship between the data bus and the DAC clock
is internally optimized. Therefore, data arrival propagation
delays and concern about data transitions near the sampling
instant are eliminated.
Synchronization is automatically enabled upon reset. After data
arrives and synchronization is achieved, the sync logic contin-
uously monitors itself so that automatic adjustments are made if
phase drifts occur over time and/or temperature.
Note that the sync function and operation of the sync logic
block are transparent, automatic, and ongoing. No programming
is required. For applications where it is useful, however, the
following programmable control is provided.
SYNC Operating States
The sync logic can operate in one of three possible modes. The
default mode is fully automatic.
Fully automatic synchronization is accomplished by demulti-
plexing the incoming data stream into four channels, each
containing every fourth data-word. Data-words are present for
four DAC clock cycles. Data is remultiplexed by sampling each
channel with the optimum DAC clock cycle.
Initial synchronization is first established through a hardware
reset. This also fully enables the synchronization logic to mon-
itor and resynchronize, as necessary. The AD9726 resynchro-
nizes only if conditions change enough to alter the phase
between the data bus and the DAC clock by more than one full
clock cycle. In this event, an internal alarm occurs and is
followed by an automatic update. During resynchronization,
two data-words are typically lost or repeated.
In addition to fully automatic mode, two semi-automatic modes
are available.
Sync Manual Mode
In fully automatic mode, the AD9726 both detects when a
resynchronization is necessary and initiates an update. In
manual mode, automatic updating is disabled. Enable manual
mode by setting the SYNCMAN bit in SPI Register 0x02.
In manual mode, the sync logic still monitors incoming data
and the DAC clock, but it indicates the need for an update by
asserting the SYNCALRM bit (Bit 0 in Register 0x02). In this
mode, the user is expected to regularly poll the SYNCALRM
bit. When this bit is read back high, the user can issue a manual
sync update also by asserting the SYNCUPD bit (Bit 1) in SPI
Register 0x02.
SYNCALRM does not indicate that data is being lost but that
conditions are close to the point where data may be lost. The
AD9726
Rev. B | Page 23 of 24
sync logic should be resynchronized by asserting SYNCUPD
at the next convenient time.
In manual mode, users can choose when to update the sync
logic. When operating with burst data, issuing a sync update
between active bursts updates the system without risking the
loss of any data. In fact, because SYNCUPD always forces a
resynchronization regardless of operational mode, even users in
fully automatic mode can reduce the possibility of data loss by
occasionally forcing a sync update during idle activity.
If either the data clock or the DAC clock is interrupted for any
reason, a SYNCUPD should always be executed to ensure that
data bus and DAC clock phase alignment remains optimized.
SYNC External Mode
Going beyond manual mode, sync external mode offers a
greater level of control and can be useful if multiple DAC
channels are employed in an application. Enable sync external
mode by asserting the SYNCEXT bit (Bit 5) in SPI Register
0x16. Manual mode must also be enabled.
The four channels into which each incoming data-word is
multiplexed are called quadrants. In any mode, the current
quadrant value can always be read back via SYNCOUT (Bits
[1:0] of SPI Register 0x15). At sync update, the logic chooses the
optimal quadrant and refreshes the value of SYNCOUT.
It is also possible to enter a value into SYNCIN (Bits [4:3] of SPI
Register 0x16). When external mode is enabled, the logic oper-
ates as expected, except that the quadrant value in SYNCIN is
used following an update. This can be used to align delays
between multiple device outputs.
Operating With SPI Disabled
If the SPI_DIS pin is connected high to ADVDD and the SPI is
disabled, the sync logic is placed into manual mode.
SYNCALRM status can then be monitored in hardware via the
unused SPI pin SDO (54), and SYNCUPD requests can be
entered in hardware via the unused SPI pin SCLK (56). If these
two pins are connected together, fully automatic sync operation
can be achieved.

AD9726BSVZ

Mfr. #:
Manufacturer:
Analog Devices Inc.
Description:
Digital to Analog Converters - DAC 16-Bit 400 MSPS
Lifecycle:
New from this manufacturer.
Delivery:
DHL FedEx Ups TNT EMS
Payment:
T/T Paypal Visa MoneyGram Western Union

Products related to this Datasheet