forked from M-Labs/artiq
coredevice/adf5356: port to NAC3
This commit is contained in:
parent
519e3d64d8
commit
e45c194c49
|
@ -8,25 +8,36 @@ on Mirny-style prefixed SPI buses.
|
|||
# https://www.analog.com/media/en/technical-documentation/data-sheets/ADF5355.pdf
|
||||
# https://www.analog.com/media/en/technical-documentation/user-guides/EV-ADF5355SD1Z-UG-1087.pdf
|
||||
|
||||
from numpy import int32, int64
|
||||
# NAC3TODO from math import floor, ceil
|
||||
|
||||
from artiq.language.core import kernel, portable, delay
|
||||
from artiq.language.core import nac3, KernelInvariant, kernel, portable, round64
|
||||
from artiq.language.units import us, GHz, MHz
|
||||
from artiq.language.types import TInt32, TInt64
|
||||
from artiq.coredevice import spi2 as spi
|
||||
from artiq.coredevice.core import Core
|
||||
from artiq.coredevice.mirny import Mirny
|
||||
from artiq.coredevice.ttl import TTLOut
|
||||
from artiq.coredevice.spi2 import *
|
||||
from artiq.coredevice.adf5356_reg import *
|
||||
|
||||
from numpy import int32, int64, floor, ceil
|
||||
|
||||
# NAC3TODO
|
||||
@portable
|
||||
def floor(x: float) -> int32:
|
||||
return 0
|
||||
@portable
|
||||
def ceil(x: float) -> int32:
|
||||
return 0
|
||||
#
|
||||
|
||||
SPI_CONFIG = (
|
||||
0 * spi.SPI_OFFLINE
|
||||
| 0 * spi.SPI_END
|
||||
| 0 * spi.SPI_INPUT
|
||||
| 1 * spi.SPI_CS_POLARITY
|
||||
| 0 * spi.SPI_CLK_POLARITY
|
||||
| 0 * spi.SPI_CLK_PHASE
|
||||
| 0 * spi.SPI_LSB_FIRST
|
||||
| 0 * spi.SPI_HALF_DUPLEX
|
||||
0 * SPI_OFFLINE
|
||||
| 0 * SPI_END
|
||||
| 0 * SPI_INPUT
|
||||
| 1 * SPI_CS_POLARITY
|
||||
| 0 * SPI_CLK_POLARITY
|
||||
| 0 * SPI_CLK_PHASE
|
||||
| 0 * SPI_LSB_FIRST
|
||||
| 0 * SPI_HALF_DUPLEX
|
||||
)
|
||||
|
||||
|
||||
|
@ -38,6 +49,7 @@ ADF5356_MAX_MODULUS2 = int32(1 << 28) # FIXME: ADF5356 has 28 bits MOD2
|
|||
ADF5356_MAX_R_CNT = int32(1023)
|
||||
|
||||
|
||||
@nac3
|
||||
class ADF5356:
|
||||
"""Analog Devices AD[45]35[56] family of GHz PLLs.
|
||||
|
||||
|
@ -49,7 +61,14 @@ class ADF5356:
|
|||
:param core_device: Core device name (default: "core")
|
||||
"""
|
||||
|
||||
kernel_invariants = {"cpld", "sw", "channel", "core", "sysclk"}
|
||||
core: KernelInvariant[Core]
|
||||
cpld: KernelInvariant[Mirny]
|
||||
channel: KernelInvariant[int32]
|
||||
sw: KernelInvariant[TTLOut]
|
||||
sysclk: KernelInvariant[float]
|
||||
regs: list[int32]
|
||||
ref_doubler: bool
|
||||
ref_divider: bool
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
|
@ -74,7 +93,7 @@ class ADF5356:
|
|||
self._init_registers()
|
||||
|
||||
@kernel
|
||||
def init(self, blind=False):
|
||||
def init(self, blind: bool = False):
|
||||
"""
|
||||
Initialize and configure the PLL.
|
||||
|
||||
|
@ -84,18 +103,20 @@ class ADF5356:
|
|||
# MUXOUT = VDD
|
||||
self.regs[4] = ADF5356_REG4_MUXOUT_UPDATE(self.regs[4], 1)
|
||||
self.sync()
|
||||
delay(1000 * us)
|
||||
self.core.delay(1000. * us)
|
||||
if not self.read_muxout():
|
||||
raise ValueError("MUXOUT not high")
|
||||
delay(800 * us)
|
||||
# NAC3TODO raise ValueError("MUXOUT not high")
|
||||
pass
|
||||
self.core.delay(800. * us)
|
||||
|
||||
# MUXOUT = DGND
|
||||
self.regs[4] = ADF5356_REG4_MUXOUT_UPDATE(self.regs[4], 2)
|
||||
self.sync()
|
||||
delay(1000 * us)
|
||||
self.core.delay(1000. * us)
|
||||
if self.read_muxout():
|
||||
raise ValueError("MUXOUT not low")
|
||||
delay(800 * us)
|
||||
# NAC3TODO raise ValueError("MUXOUT not low")
|
||||
pass
|
||||
self.core.delay(800. * us)
|
||||
|
||||
# MUXOUT = digital lock-detect
|
||||
self.regs[4] = ADF5356_REG4_MUXOUT_UPDATE(self.regs[4], 6)
|
||||
|
@ -103,7 +124,7 @@ class ADF5356:
|
|||
self.sync()
|
||||
|
||||
@kernel
|
||||
def set_att_mu(self, att):
|
||||
def set_att_mu(self, att: int32):
|
||||
"""Set digital step attenuator in machine units.
|
||||
|
||||
:param att: Attenuation setting, 8 bit digital.
|
||||
|
@ -111,11 +132,11 @@ class ADF5356:
|
|||
self.cpld.set_att_mu(self.channel, att)
|
||||
|
||||
@kernel
|
||||
def write(self, data):
|
||||
def write(self, data: int32):
|
||||
self.cpld.write_ext(self.channel | 4, 32, data)
|
||||
|
||||
@kernel
|
||||
def read_muxout(self):
|
||||
def read_muxout(self) -> bool:
|
||||
"""
|
||||
Read the state of the MUXOUT line.
|
||||
|
||||
|
@ -124,7 +145,7 @@ class ADF5356:
|
|||
return bool(self.cpld.read_reg(0) & (1 << (self.channel + 8)))
|
||||
|
||||
@kernel
|
||||
def set_output_power_mu(self, n):
|
||||
def set_output_power_mu(self, n: int32):
|
||||
"""
|
||||
Set the power level at output A of the PLL chip in machine units.
|
||||
|
||||
|
@ -132,13 +153,14 @@ class ADF5356:
|
|||
|
||||
:param n: output power setting, 0, 1, 2, or 3 (see ADF5356 datasheet, fig. 44).
|
||||
"""
|
||||
if n not in [0, 1, 2, 3]:
|
||||
raise ValueError("invalid power setting")
|
||||
if not 0 <= n <= 3:
|
||||
# NAC3TODO raise ValueError("invalid power setting")
|
||||
pass
|
||||
self.regs[6] = ADF5356_REG6_RF_OUTPUT_A_POWER_UPDATE(self.regs[6], n)
|
||||
self.sync()
|
||||
|
||||
@portable
|
||||
def output_power_mu(self):
|
||||
def output_power_mu(self) -> int32:
|
||||
"""
|
||||
Return the power level at output A of the PLL chip in machine units.
|
||||
"""
|
||||
|
@ -161,25 +183,27 @@ class ADF5356:
|
|||
self.sync()
|
||||
|
||||
@kernel
|
||||
def set_frequency(self, f):
|
||||
def set_frequency(self, f: float):
|
||||
"""
|
||||
Output given frequency on output A.
|
||||
|
||||
:param f: 53.125 MHz <= f <= 6800 MHz
|
||||
"""
|
||||
freq = int64(round(f))
|
||||
freq = round64(f)
|
||||
|
||||
if freq > ADF5356_MAX_VCO_FREQ:
|
||||
raise ValueError("Requested too high frequency")
|
||||
# NAC3TODO raise ValueError("Requested too high frequency")
|
||||
pass
|
||||
|
||||
# select minimal output divider
|
||||
rf_div_sel = 0
|
||||
while freq < ADF5356_MIN_VCO_FREQ:
|
||||
freq <<= 1
|
||||
freq <<= int64(1)
|
||||
rf_div_sel += 1
|
||||
|
||||
if (1 << rf_div_sel) > 64:
|
||||
raise ValueError("Requested too low frequency")
|
||||
# NAC3TODO raise ValueError("Requested too low frequency")
|
||||
pass
|
||||
|
||||
# choose reference divider that maximizes PFD frequency
|
||||
self.regs[4] = ADF5356_REG4_R_COUNTER_UPDATE(
|
||||
|
@ -193,7 +217,7 @@ class ADF5356:
|
|||
n_min, n_max = 75, 65535
|
||||
|
||||
# adjust reference divider to be able to match n_min constraint
|
||||
while n_min * f_pfd > freq:
|
||||
while int64(n_min) * f_pfd > freq:
|
||||
r = ADF5356_REG4_R_COUNTER_GET(self.regs[4])
|
||||
self.regs[4] = ADF5356_REG4_R_COUNTER_UPDATE(self.regs[4], r + 1)
|
||||
f_pfd = self.f_pfd()
|
||||
|
@ -207,7 +231,8 @@ class ADF5356:
|
|||
)
|
||||
|
||||
if not (n_min <= n <= n_max):
|
||||
raise ValueError("Invalid INT value")
|
||||
# NAC3TODO raise ValueError("Invalid INT value")
|
||||
pass
|
||||
|
||||
# configure PLL
|
||||
self.regs[0] = ADF5356_REG0_INT_VALUE_UPDATE(self.regs[0], n)
|
||||
|
@ -221,10 +246,10 @@ class ADF5356:
|
|||
|
||||
self.regs[6] = ADF5356_REG6_RF_DIVIDER_SELECT_UPDATE(self.regs[6], rf_div_sel)
|
||||
self.regs[6] = ADF5356_REG6_CP_BLEED_CURRENT_UPDATE(
|
||||
self.regs[6], int32(floor(24 * f_pfd / (61.44 * MHz)))
|
||||
self.regs[6], floor(24. * float(f_pfd) / (61.44 * MHz))
|
||||
)
|
||||
self.regs[9] = ADF5356_REG9_VCO_BAND_DIVISION_UPDATE(
|
||||
self.regs[9], int32(ceil(f_pfd / 160e3))
|
||||
self.regs[9], ceil(float(f_pfd) / 160e3)
|
||||
)
|
||||
|
||||
# commit
|
||||
|
@ -236,21 +261,21 @@ class ADF5356:
|
|||
Write all registers to the device. Attempts to lock the PLL.
|
||||
"""
|
||||
f_pfd = self.f_pfd()
|
||||
delay(200 * us) # Slack
|
||||
self.core.delay(200. * us) # Slack
|
||||
|
||||
if f_pfd <= 75.0 * MHz:
|
||||
if f_pfd <= round64(75.0 * MHz):
|
||||
for i in range(13, 0, -1):
|
||||
self.write(self.regs[i])
|
||||
delay(200 * us)
|
||||
self.core.delay(200. * us)
|
||||
self.write(self.regs[0] | ADF5356_REG0_AUTOCAL(1))
|
||||
else:
|
||||
# AUTOCAL AT HALF PFD FREQUENCY
|
||||
|
||||
# calculate PLL at f_pfd/2
|
||||
n, frac1, (frac2_msb, frac2_lsb), (mod2_msb, mod2_lsb) = calculate_pll(
|
||||
self.f_vco(), f_pfd >> 1
|
||||
self.f_vco(), f_pfd >> int64(1)
|
||||
)
|
||||
delay(200 * us) # Slack
|
||||
self.core.delay(200. * us) # Slack
|
||||
|
||||
self.write(
|
||||
13
|
||||
|
@ -273,7 +298,7 @@ class ADF5356:
|
|||
)
|
||||
self.write(1 | ADF5356_REG1_MAIN_FRAC_VALUE(frac1))
|
||||
|
||||
delay(200 * us)
|
||||
self.core.delay(200. * us)
|
||||
self.write(ADF5356_REG0_INT_VALUE(n) | ADF5356_REG0_AUTOCAL(1))
|
||||
|
||||
# RELOCK AT WANTED PFD FREQUENCY
|
||||
|
@ -285,7 +310,7 @@ class ADF5356:
|
|||
self.write(self.regs[0] & ~ADF5356_REG0_AUTOCAL(1))
|
||||
|
||||
@portable
|
||||
def f_pfd(self) -> TInt64:
|
||||
def f_pfd(self) -> int64:
|
||||
"""
|
||||
Return the PFD frequency for the cached set of registers.
|
||||
"""
|
||||
|
@ -295,35 +320,35 @@ class ADF5356:
|
|||
return self._compute_pfd_frequency(r, d, t)
|
||||
|
||||
@portable
|
||||
def f_vco(self) -> TInt64:
|
||||
def f_vco(self) -> int64:
|
||||
"""
|
||||
Return the VCO frequency for the cached set of registers.
|
||||
"""
|
||||
return int64(
|
||||
self.f_pfd()
|
||||
return round64(
|
||||
float(self.f_pfd())
|
||||
* (
|
||||
self.pll_n()
|
||||
+ (self.pll_frac1() + self.pll_frac2() / self.pll_mod2())
|
||||
/ ADF5356_MODULUS1
|
||||
)
|
||||
float(self.pll_n())
|
||||
+ (float(self.pll_frac1() + self.pll_frac2()) / float(self.pll_mod2()))
|
||||
/ float(ADF5356_MODULUS1)
|
||||
)
|
||||
)
|
||||
|
||||
@portable
|
||||
def pll_n(self) -> TInt32:
|
||||
def pll_n(self) -> int32:
|
||||
"""
|
||||
Return the PLL integer value (INT) for the cached set of registers.
|
||||
"""
|
||||
return ADF5356_REG0_INT_VALUE_GET(self.regs[0])
|
||||
|
||||
@portable
|
||||
def pll_frac1(self) -> TInt32:
|
||||
def pll_frac1(self) -> int32:
|
||||
"""
|
||||
Return the main fractional value (FRAC1) for the cached set of registers.
|
||||
"""
|
||||
return ADF5356_REG1_MAIN_FRAC_VALUE_GET(self.regs[1])
|
||||
|
||||
@portable
|
||||
def pll_frac2(self) -> TInt32:
|
||||
def pll_frac2(self) -> int32:
|
||||
"""
|
||||
Return the auxiliary fractional value (FRAC2) for the cached set of registers.
|
||||
"""
|
||||
|
@ -332,7 +357,7 @@ class ADF5356:
|
|||
) | ADF5356_REG2_AUX_FRAC_LSB_VALUE_GET(self.regs[2])
|
||||
|
||||
@portable
|
||||
def pll_mod2(self) -> TInt32:
|
||||
def pll_mod2(self) -> int32:
|
||||
"""
|
||||
Return the auxiliary modulus value (MOD2) for the cached set of registers.
|
||||
"""
|
||||
|
@ -341,14 +366,14 @@ class ADF5356:
|
|||
) | ADF5356_REG2_AUX_MOD_LSB_VALUE_GET(self.regs[2])
|
||||
|
||||
@portable
|
||||
def ref_counter(self) -> TInt32:
|
||||
def ref_counter(self) -> int32:
|
||||
"""
|
||||
Return the reference counter value (R) for the cached set of registers.
|
||||
"""
|
||||
return ADF5356_REG4_R_COUNTER_GET(self.regs[4])
|
||||
|
||||
@portable
|
||||
def output_divider(self) -> TInt32:
|
||||
def output_divider(self) -> int32:
|
||||
"""
|
||||
Return the value of the output A divider.
|
||||
"""
|
||||
|
@ -398,7 +423,7 @@ class ADF5356:
|
|||
|
||||
# single-ended reference mode is recommended
|
||||
# for references up to 250 MHz, even if the signal is differential
|
||||
if self.sysclk <= 250 * MHz:
|
||||
if self.sysclk <= 250.*MHz:
|
||||
self.regs[4] |= ADF5356_REG4_REF_MODE(0)
|
||||
else:
|
||||
self.regs[4] |= ADF5356_REG4_REF_MODE(1)
|
||||
|
@ -440,7 +465,7 @@ class ADF5356:
|
|||
|
||||
# charge pump bleed current
|
||||
self.regs[6] |= ADF5356_REG6_CP_BLEED_CURRENT(
|
||||
int32(floor(24 * self.f_pfd() / (61.44 * MHz)))
|
||||
floor(24. * float(self.f_pfd()) / (61.44 * MHz))
|
||||
)
|
||||
|
||||
# direct feedback from VCO to N counter
|
||||
|
@ -484,7 +509,7 @@ class ADF5356:
|
|||
)
|
||||
|
||||
self.regs[9] |= ADF5356_REG9_VCO_BAND_DIVISION(
|
||||
int32(ceil(self.f_pfd() / 160e3))
|
||||
ceil(float(self.f_pfd()) / 160e3)
|
||||
)
|
||||
|
||||
# REG10
|
||||
|
@ -513,39 +538,39 @@ class ADF5356:
|
|||
self.regs[12] = int32(0x15FC)
|
||||
|
||||
@portable
|
||||
def _compute_pfd_frequency(self, r, d, t) -> TInt64:
|
||||
def _compute_pfd_frequency(self, r: int32, d: int32, t: int32) -> int64:
|
||||
"""
|
||||
Calculate the PFD frequency from the given reference path parameters
|
||||
"""
|
||||
return int64(self.sysclk * ((1 + d) / (r * (1 + t))))
|
||||
return round64(self.sysclk * (float(1 + d) / float(r * (1 + t))))
|
||||
|
||||
@portable
|
||||
def _compute_reference_counter(self) -> TInt32:
|
||||
def _compute_reference_counter(self) -> int32:
|
||||
"""
|
||||
Determine the reference counter R that maximizes the PFD frequency
|
||||
"""
|
||||
d = ADF5356_REG4_R_DOUBLER_GET(self.regs[4])
|
||||
t = ADF5356_REG4_R_DIVIDER_GET(self.regs[4])
|
||||
r = 1
|
||||
while self._compute_pfd_frequency(r, d, t) > ADF5356_MAX_FREQ_PFD:
|
||||
while self._compute_pfd_frequency(r, d, t) > int64(ADF5356_MAX_FREQ_PFD):
|
||||
r += 1
|
||||
return int32(r)
|
||||
return r
|
||||
|
||||
|
||||
@portable
|
||||
def gcd(a, b):
|
||||
while b:
|
||||
def gcd(a: int64, b: int64) -> int64:
|
||||
while b != int64(0):
|
||||
a, b = b, a % b
|
||||
return a
|
||||
|
||||
|
||||
@portable
|
||||
def split_msb_lsb_28b(v):
|
||||
return int32((v >> 14) & 0x3FFF), int32(v & 0x3FFF)
|
||||
def split_msb_lsb_28b(v: int32) -> tuple[int32, int32]:
|
||||
return (v >> 14) & 0x3FFF, v & 0x3FFF
|
||||
|
||||
|
||||
@portable
|
||||
def calculate_pll(f_vco: TInt64, f_pfd: TInt64):
|
||||
def calculate_pll(f_vco: int64, f_pfd: int64) -> tuple[int32, int32, tuple[int32, int32], tuple[int32, int32]]:
|
||||
"""
|
||||
Calculate fractional-N PLL parameters such that
|
||||
|
||||
|
@ -558,25 +583,23 @@ def calculate_pll(f_vco: TInt64, f_pfd: TInt64):
|
|||
:param f_pfd: PFD frequency
|
||||
:return: ``(n, frac1, (frac2_msb, frac2_lsb), (mod2_msb, mod2_lsb))``
|
||||
"""
|
||||
f_pfd = int64(f_pfd)
|
||||
f_vco = int64(f_vco)
|
||||
|
||||
# integral part
|
||||
n, r = int32(f_vco // f_pfd), f_vco % f_pfd
|
||||
|
||||
# main fractional part
|
||||
r *= ADF5356_MODULUS1
|
||||
r *= int64(ADF5356_MODULUS1)
|
||||
frac1, frac2 = int32(r // f_pfd), r % f_pfd
|
||||
|
||||
# auxiliary fractional part
|
||||
mod2 = f_pfd
|
||||
|
||||
while mod2 > ADF5356_MAX_MODULUS2:
|
||||
mod2 >>= 1
|
||||
frac2 >>= 1
|
||||
while mod2 > int64(ADF5356_MAX_MODULUS2):
|
||||
mod2 >>= int64(1)
|
||||
frac2 >>= int64(1)
|
||||
|
||||
gcd_div = gcd(frac2, mod2)
|
||||
mod2 //= gcd_div
|
||||
frac2 //= gcd_div
|
||||
|
||||
return n, frac1, split_msb_lsb_28b(frac2), split_msb_lsb_28b(mod2)
|
||||
return n, frac1, split_msb_lsb_28b(int32(frac2)), split_msb_lsb_28b(int32(mod2))
|
||||
|
|
Loading…
Reference in New Issue