Compensating for Bondwire Inductance
Although the fabrication procedure used results in
the shortest bondwire lengths, which are practical,
there is still an associated inductance of around
0.3nH per bond, for 0.001” diameter wire. At lower
frequencies the reactance this represents is very low
and can be ignored. At mm-wave frequencies even
such low inductances as this can cause significant
performance degradation. Figure 3 shows a plot of
the simulated insertion loss and match of a 0.3nH
bondwire versus frequency, in a 50Ω system. The return loss of the bondwire has fallen to below 10dB by
18GHz and to below 7dB by 28GHz with an associated insertion loss of more than 1dB.
It is clear that simply accepting the performance degradation caused by a 0.3nH bondwire at mm-wave
frequencies is not acceptable. There are three options for resolving the problem:
1. Reduce the inductance by using multiple parallel bondwires or gold tape.
2. Use ICs which have been designed to accommodate a 0.3nH inductor at all RF ports.
3. Incorporate the inductance into a low pass filter structure.
Option 1 is not the best approach for low cost, high volume use. Multiple bonds increase assembly time
and require a larger, non-standard bondpad (with more parasitic shunt capacitance) and tape bonding has
a significant cost penalty. Option 2 is viable but requires all ICs to be designed for a specific assembly
process. Also the RF On Wafer (RFOW) measured performance of the ICs will differ significantly from
the in-circuit
performance as the RF
port bondwire
inductance, which the
circuit’s performance
has been optimised to
include, would be
missing. Option 3 allows
ICs designed for best
RFOW performance to
be used. So long as the
bondwire inductance is
low enough, it can be
absorbed into a practical,
low pass filter structure,
such as that shown in
Figure 4. The printed
open circuit stubs act as a shunt capacitance and the narrow (high impedance) series microstrip line serves
as an inductance to complete the third order low pass filter structure.Figure 5 shows the simulated performance of a low
pass filter design, which uses this technique to
incorporate a 0.3nH bondwire inductance. It has
been optimised for use up to 30GHz and improved
return loss and insertion loss, as compared to the
simple series bondwire case, is evident above
10GHz.
Downconverter Measured Performance
A downconverter MCM has been designed and
fabricated using the methods described above. It
operates over an RF frequency range of 27.5 to
29.5GHz, with an IF of 4GHz. The LO input is in
the range 11.75 to 12.75GHz, as the downconverter
utilises a sub-harmonic mixer which accepts a half
frequency LO. All components used are
commercially available. A block diagram of the
complete downconverter is shown in Figure 6.
The LNA and the mixer are both 0.25μm gate length
Pseudomorphic High Electron Mobility Transistor
(PHEMT) MMICs, which are used in bare die form.
Image filtering is realised using a five element, printed
coupled line filter [1], whilst the low pass IF filter is a
printed stub design. It is strictly a band-stop filter,
which rejects the half LO output of the mixer, which
can be particularly high for sub-harmonic mixers. An
inexpensive SMT component is used to realise the IF
amplifier, with a network of 0402 passives and printed
stubs to flatten the gain versus frequency response.
In addition to the complete downconverter, a number
of sub-circuits were fabricated on the same tile for
diagnostic purposes. Figure 7 shows the measured
performance of a sub-circuit comprising the IF low pass filter and IF amplifier. It exhibits a gain of
16dB at 4GHz and a rejection of over 55dB for
the 11.75 to 12.75GHz half LO frequency range.
The conversion gain, versus frequency of the
complete downconverter is around 23dB, as
shown in Figure 9. Image rejection is over 35dB,
across the band.
A plot of the IF port output spectrum for an RF
input of -40dBm is shown in Figure 10. The subharmonic
mixer contains a half LO amplifier and
the level of unfiltered half LO at the IF output of
the mixer is around +4dBm. This is quite
significant and is the reason
for the bandstop nature of
the IF filter. The half LO
level at the output of the
entire downconverter is
-39dBm, having been
substantially attenuated by
the IF filter. The quarter LO
products are a result of
quarter LO output from the
signal source used to drive
the LO. If this frequency
component is present in the
end system, a simple high
pass filter on the LO port of
the mixer can be used to
provide attenuation.
Upconverter Measured Performance
Like the downconverter, the upconverter adopts a heterodyne architecture and uses only commercially
available parts. Figure 11 shows a block diagram of the upconverter, which uses the same sub-harmonic
mixer as the downconverter. All of the transmit chain RF amplifiers are 0.25μm gate length PHEMT
MMICs, in bare die form.
The IF amplifier and low pass filter are similar
to those in the receiver, although an amplifier
with higher intermodulation performance is
used. Upconverters for LMDS systems, operating
in the 28GHz band, are likely to use nonconstant
envelope modulation schemes and good
transmitter linearity will be important. Some
system developers are also considering higher
order modulation schemes, such as 16-QAM
(Quadrature Amplitude Modulation), where the
requirements for linearity will be even more
stringent in order to preserve modulation
fidelity.
The unwanted sideband output from the mixer
(LO-IF) will be at around the same level as the
wanted RF signal (RF = LO+IF). The LO output
will also be at a significant level and these, together with any other unwanted spurious outputs must be
heavily attenuated prior to final amplification. Two five element coupled line band pass filters are used,
one either side of the variable gain pre-driver,
Figure 12 shows the measured performance of a
sub-circuit containing this structure. The image
and LO rejection is over 70dB.The variable gain
amplifier has a gain control range of over 15dB
and can be used to adjust the post mixer gain to
optimise linearity or compensate for gain
variation with frequency or part to part. The
overall conversion gain of the entire
upconverter is around 43dB. Figure 13 shows
the measured power transfer characteristics
with a 1dB gain compressed output power level
of +23dBm
Conclusions
A low cost process for manufacturing mm-wave modules, which is suitable for automation in volume
manufacture, has been developed. A PTFE composite substrate is used, with mm-wave MMICs and SMT
components assembled in a single process step. Two 27.5 to 29.5GHz MCMs, a downconverter and an
upconverter, have been designed, manufactured and measured. The downconveter has a conversion gain
of 23dB with 35dB of image rejection. The upconverter has a conversion gain of 43dB with a 1dB gain
compressed output power of +23dBm. All of the components used are commercially available. |
