Performance requirement and lessons learnt of LTE terminal---transmitter part

Transmitter Output Power
EVM
ACLR

By Criterion 3
Transmitter
Output Power

By Criterion 4
Specification
 The LTE specified maximum output power：23 dBm with a
tolerance of ± 2 dB [1,6]. Nevertheless, the WCDMA specified
24 dBm with a tolerance of +1/−3 dB [1].

By Criterion 5
key parameters
 The overall LTE key parameters are shown below [2]:

By Criterion
bits per symbol
 LTE uses QPSK, 16 QAM, and 64 QAM for uplink
transmission, whereas WCDMA makes use of BPSK for
uplink transmission. The overall bits per symbol table for
these modulation types is summarized as below [3]:

By Criterion
bits per symbol
 BPSK = 𝟐 𝟏
sym𝐛𝐨𝐥𝐬, 𝐛𝐢𝐭𝐬 𝐩𝐞𝐫 𝐬𝐲𝐦𝐛𝐨𝐥 = 𝟏
QPSK = 𝟐 𝟐
sym𝐛𝐨𝐥𝐬, 𝐛𝐢𝐭𝐬 𝐩𝐞𝐫 𝐬𝐲𝐦𝐛𝐨𝐥 = 𝟐
64QAM = 𝟐 𝟔
sym𝐛𝐨𝐥𝐬, 𝐛𝐢𝐭𝐬 𝐩𝐞𝐫 𝐬𝐲𝐦𝐛𝐨𝐥 = 𝟔

By Criterion
PAR (Peak -to-Average-Ratio)
 Higher order modulation achieves higher data rate at the
expense of higher PAR (Peak -to-Average-Ratio), which
requires more back-off to retain linearity [1,5].

By Criterion
 Hence, there are average power and peak power on the
display simultaneously.

By Criterion
 SC-FDMA is used for LTE uplink transmission to reduce PAR
[5]. Nevertheless, LTE still has higher PAR in uplink
modulation than WCDMA[1]. Thus, that’s why maximum
output power of LTE is 23 dBm, but 24 dBm of WCDMA.

By Criterion
MPR (Maximum Power Reduction)
 Furthermore, MPR (Maximum Power Reduction) has been
introduced in LTE to take account of the higher PAR of 16-
QAM modulation and RB (Resource Block) allocation [1,6].
That’s why the output power is lower under full RBs.

By Criterion
Headroom
 LTE power control ranges from -40 dBm to 23 dBm. [1]
Take SKYWORKS SKY77645-11 for example, its maximum
power in LTE B7 is 28.5 dBm, which indicates the MMPA post-
loss should be less than 5.5 dB.
 Of course, the post-loss is the smaller the better since more
headroom leads to better linearity.

By Criterion
Headroom
 Moreover, the output power is relevant for temperature and
voltage, so the compensation should be done to ensure
consistent output power under various conditions.
 The worst case is (low temperature + high voltage), which
makes output power rise dramatically, thereby aggravating
TX performance.

By Criterion
Headroom
 As shown below, the saturated power for low gain mode is
26.9 dBm, and which for high gain mode is 29 dBm. In other
words, both low gain mode and high gain mode can achieve
23 dBm. Nevertheless, high gain mode has more headroom
(29 dBm – 23 dBm = 6 dB) than low gain mode (26.9 dBm – 23
dBm = 3.9 dB). So incorrect gain mode also causes linearity
issues, such as EVM.

By Criterion
FBRX
 As for power control, Qualcomm makes use of FBRX
(Feedback Receiver) method:

By Criterion
FBRX
 The output power is usually coupled back to transceiver by
means of the coupler integrated in ASM (Antenna Switch
Module). Take SKYWORKS SKY77912-11 for example [9]
 Thus, the overall process is closed loop, thereby adjusting
output power by the feedback power to retain accuracy.

By Criterion
Lesson learnt_1: B28 Minimum power issue
 While doing minimum power measurement, B27 can achieve
the power level less than -30 dBm, but B28 can NOT.
 B27/B28 share the identical transmitting path. In other words,
the issue is not due to hardware issue since B27 is normal.

By Criterion
 While doing calibration, with the identical RGI (RF Gain
Index), there is approximately 10 dB gap between B27 and
B28, as shown below [10]:

By Criterion
 Nevertheless, with the same RGI, there is no difference
between B27 and B28 output power at transceiver output port.
This indicates that the issue is not related to transceiver.

By Criterion
 As shown below, B27 and B28 should be configured LB and
VLB respectively. Otherwise, there may be something wrong
in transmitting performance, such as minimum power. So the
root cause is incorrect configuration for B28.

By Criterion
Lesson learnt_2: Band 2 low channel Max power
 As shown below, while doing B2 maximum power
measurement, it’s too high in low channel, but normal in mid
and high channels.
 From calibration log, it is apparent that the low channels
have lower HDET (High Power Detector) value than mid and
high ones.

By Criterion
 As shown below, the measured maximum power at connector
should be 23 dBm. We assume the loss of duplexer is
approximately 3 dB, so the output power at MMPA is 26 dBm.
 Because the coupler is integrated into MMPA and the
coupling factor is 23 dB, the feedback power is 3 dBm, which
is higher than the handling capacity of feedback LNA

By Criterion
 As shown below, with a strong input signal, the gain of LNA
reduces [11]. That’s why the low channels have lower HDET
value than mid and high ones.

By Criterion
 As for mid and high channels, perhaps the frequency
response of FBRX path is as shown below:
 Thus, thanks to higher path loss, the feedback power doesn’t
make feedback LNA saturate in mid and high channels. This
explains why HDET value increases as channel number
increases reasonably.

By Criterion
 Thanks to lower HDET value in low channels, the baseband
block misunderstands that the output power is too low,
thereby increasing RGI and resulting in too high output
power in low channels. Therefore. In modern LTE terminal
design, the coupler is integrated into ASM, instead of MMPA.
So the root cause is too high feedback power in low channel.

By Criterion
Lesson learnt_3: All bands too high maximum power
 During factory manufacturing phase, one board failed thanks
to too large power. Compared to good board, the value of
stored NV item (FBRX_Gain_Value) is larger than good board.

By Criterion
Lesson learnt_3: All bands too high maximum power
 Since there is soldering issue in DC block, thereby resulting
in large loss in FBRX path. With weak input signal, the LNA
switches to high gain mode to lower the overall noise figure
to achieve acceptable BER. Additionally, the baseband block
misunderstands that the output power is too low, thereby
increasing RGI and resulting in too high output power.
So the root cause is too low feedback power
in all channels.

By Criterion
Lesson learnt_4: Band 38 too high maximum power
 As shown below, B38 and B41 share the same transmitting
path, but the maximum output power of B38 is higher than
B41 approximately 2 dB.

By Criterion
 From calibration log, B38 has lower calibration power than
B41. Thus, this issue is related to FBRX.

By Criterion
 Check the coupler configuration, B38 is configured LB
instead of HB. As shown below, the coupler has larger
coupling factor in LB (27 dB) than in HB (22 dB).

By Criterion
 It means that B38 has larger loss in FBRX path than B41.
With weak input signal, the baseband block misunderstands
that the output power is too low, thereby increasing RGI and
resulting in too high output power in B38.

By Criterion
 Similarly, if B5 is configured HB, as shown below:
Since the coupling factor of HB is less than LB approximately
5 dB, the B5 feedback power will be higher than expectation,
thereby making feed-back LNA saturate. With low output
power level from feed-back LNA(due to gain reduction), the
baseband block may misunderstand the output power is too
low, thereby increasing RGI and resulting in too high output
power.

By Criterion
 The layout is as shown below:

By Criterion
 The impedance may alter while the shielding cover is on the
co-planar ground since the medium is already NOT air.

By Criterion
 B7 is high band, which is more sensitive to impedance
variation than mid- and low bands. Thus, the impedance
mismatch in FBRX path results in high mismatch loss,
thereby leading to too high power due to compensation
mechanism.
 Because FBRX affects numerous bands, please lay the trace
in inner layer to obtain better protection. Otherwise, output
power, EVM, and ILPC (Inner Loop Power Control) may fail.

By Criterion
Lesson learnt_6: LTE B40 high channel too low max power
 As mentioned earlier, for the output power, the compensation
should be done to ensure consistent output power under
various temperature and frequency. Similarly, compensation
should also be done properly for FBRX because this affects
output power.

By Criterion
Lesson learnt_6: LTE B40 high channel too low max power
 As shown below, the FBRX high channel compensation is
exceeding other channels, thereby making baseband block
misunderstand the output power is too high.
Thanks to compensation mechanism,
the RGI will become lower, thereby causing
too low output power.
 Similarly, if the compensation is much less
than other channels, the baseband block
may misunderstand the output power is
too low. Thanks to compensation mechanism,
the RGI will become higher, thereby causing
too high output power.

By Criterion
How does FBRX affect output power?
 Hence, these aforementioned power issues tell us that :
 Too much loss in FBRX path => Too high output power
 Too high feedback power => Too high output power
 Exceeding compensation in FBRX => Too low output power
 Insufficient compensation in FBRX => Too high outputpower
 In terms of hardware, check FBRX whenever output power is
too high.

By Criterion 40
Introduction
 As shown below, there are numerous test items for signal
quality [6].

By Criterion 41
Introduction
 EVM (Error Vector Magnitude) is as depicted below [6]:

By Criterion 42
Introduction
 EVM is a vector in the I-Q plane between the ideal
constellation point and the practical point received by the
receiver. In other words, it is the difference between actual
received symbols and ideal symbols. EVM is low if the
actual received symbols are very close to ideal symbols,
and vice versa.
 In time domain, the timing error in waveform is called “jitter”,
which generates phase error in the modulation constellation,
thereby contributing to EVM [1].

By Criterion 43
Phase Noise
 Let’s look at the effect of LO phase noise from another point
of view.
 According to this relationship [12], it is apparent that the
EVM is inversely proportional to SNR. As shown below, a LO
with high phase noise leads to the reduction in RF signal
SNR, thereby aggravating EVM performance.

By Criterion 44
Phase Noise
 As shown below, take Qualcomm WTR4905 for example, the
crystal generates a 19.2 MHz sine waveform to produce a
19.2 MHz square-wave XO signal from PMIC to transceiver
[13]. Thus, a crystal with high phase noise contributes to the
phase noise of LO

By Criterion 45
Crystal Oscillator
 Nevertheless, cellular technology usually shares one
identical crystal with GPS technology. If GPS performance is
acceptable (i.e., CNR = 38 ~ 40 dB), it indicates that the
crystal is innocent since GPS performance is more sensitive
to crystal performance thanks to its extremely weak
received signal. Poor EVM
Poor GPS CNR ?
Check XO
XO is innocent
Yes
No

By Criterion 46
Crystal Oscillator
 The digital square-wave signals would corrupt analog sine-
wave signal; sufficient isolation is highly recommended [13].
The layout shown below is a bad example since the isolation
is not enough.

By Criterion 47
VCO
 Furthermore, keep-out areas on PCB top layer is necessary
for transceiver since these areas is related to VCO, which is
sensitive to parasitic effect. Otherwise, the parasitic effect
may aggravate VCO phase noise [13, 30].

By Criterion 48
OFDM
 LTE makes use of OFDM (Orthogonal Frequency Division
Multiplexing) modulation [5], which is sensitive to phase
error and/or frequency offset [14].

By Criterion 49
VCO Pulling
 As shown below, in direct-conversion transmitter
architecture, perhaps an appreciable fraction of the PA
output couples to the LO trough substrate, and PCB traces
etc. [16].

By Criterion 50
VCO Pulling
 With the presence of injection pulling, the LO output
waveform is as shown below [16]:

By Criterion 51
VCO Pulling
 As mentioned earlier, the timing error in waveform generates
phase error in the modulation constellation, thereby
contributing to EVM, and OFDM modulation is sensitive to
phase error. Thus, for a LTE system, especially direct-
conversion transceiver architecture (since RF frequency
almost equals to LO frequency), oscillator pulling should be
avoided.

By Criterion 52
Lesson learnt_7: EVM issue due to shielding can
 As shown below, if someone presses the shielding can with
finger, the EVM performance is good, and vice versa.

By Criterion 53
 That’s because PA and transceiver blocks are placed in the
identical shielding space. The PA couples its strong RF
energy onto the shielding can. An appreciable fraction of the
PA output couples to the VCO trough shielding can by
means of reflection, thereby making VCO pulling happen
and aggravating EVM performance [13,18].

By Criterion 54
 Nevertheless, if someone presses the shielding can, this
action reinforces the grounding of the shielding can, thereby
eliminating reflection and VCO pulling [13].

By Criterion 55
 Thus, the PA and transceiver blocks should be placed in
separated shielding areas individually to avoid oscillator
pulling [30].

By Criterion 56
Mismatch between transceiver and PA
 Moreover, for typical designs, as the PA output exceeds 0
dBm, injection pulling may prove severe [16].
 Nevertheless, take Qualcomm SDR660 for example [19], the
output power from transceiver exceeds 0 dBm.

By Criterion 57
Mismatch between transceiver and PA
 In other words, not only PA output causes injection pulling,
but also transceiver output does. If the impedance between
transceiver and PA is NOT 50 Ohm, the reflection thanks to
mismatch may cause injection pulling [13].

By Criterion 58
Lesson learnt_8: B41 EVM issue due to RX path
 Modern MMPAs adopt fully programmable MIPI (Mobile
Industry Processor Interface) control, which can easily
enable more than one RF path simultaneously [8,28]. B41 is
TDD, receiver path should be off while transmitter is
operating. Because improper value is written into PA register,
which makes receiver path enable (marked as red circle)
while transmitter is operating.

By Criterion 59
Lesson learnt_8: B41 EVM issue due to RX path
 Thus, a fraction of PA output power leaks to transceiver
through receiver path, thereby causing VCO pulling. Unlike
FDD system, TDD system does NOT have duplexer to
suppress TX leakage. Thus, the VCO pulling proves severe
and causes EVM issue.

By Criterion 60
PA Nonlinearity
 Additionally, the PA nonlinearity, AM-AM and AM-PM
distortion, causes the amplitude error and phase error on
the output signal and then has a contribution to EVM [1,20-
21].

By Criterion 61
PA Nonlinearity
 As shown above, the AM-AM and AM-PM distortion
increases as the output power is increased [1]. Especially,
as mentioned earlier, OFDM-based transmitter is sensitive to
phase error [20]. Thus, the saturated output power should
be high enough to posses more headroom, thereby causing
more back-off and better linearity [1]. This is especially
important for TDD (Time Division Duplexing) bands such as
B38/B40/B41 since PA is pulsed on and off during usage.
This kind of dynamic mode has worse linearity performance
than static mode (i.e., FDD bands) [26].

By Criterion 62
PA Nonlinearity
 The modulation scheme and maximum output power per
band configurations for a MMPA is as shown below [1]:

By Criterion 63
DPD
 Usually, there are three ways to improve PA linearity. First,
do DPD (Digital Pre-Distortion), which is a method
universally adopted and employed in wireless cellular
industry [1,5, 22].

By Criterion 64
Load-pull
 Secondly, tune PA output impedance to transform the load
impedance (usually 50 Ohm) into the desired one [1].
Reference
Plane PA Output pin

By Criterion 65
Load-pull
 As shown below,
the blue contour
represents efficiency (%)
and red contour
represents
saturated power [23].
It is apparently that
there is a trade-off between
efficiency and linearity.
Thus, you need to optimize
saturated power
at the expense of efficiency.

By Criterion 66
Envelope Tracking
 Third, tune the PA voltage supply [1]. ET (Envelope-tracking)
is a technique for improving the energy efficiency of PA. The
traditional DC-DC converter supplying (usually from PMIC
directly) is replaced by a highly agile ET power supply
modulating the power supply of the PA. It means that the PA
is always operating in a highly efficient compressed state
[24].

By Criterion 67
Envelope Tracking
 As mentioned earlier, there is a trade-off between linearity
and efficiency. Thus, ET improves efficiency at the expense
of linearity. Conversely, you can also optimize linearity at the
expense of efficiency. That’s why you can tune the PA
voltage supply. Practically, you can tune ICQ (Quiescent
Current) point.

By Criterion 68
I/Q Signals
 Quadrature imperfections in the up-mixing can cause IQ
imbalance, which distorts constellation symbol location and
aggravates EVM performance [13,17,25].

By Criterion 69
I/Q Signals
 Excessive DC component in I/Q branches cause high level
of carrier leakage (IQ origin offset), which also distorts
constellation symbol location and aggravates EVM
performance [13].

By Criterion 70
I/Q Signals
 As shown below, “1” is just RF transmitting signal, “2” is
carrier leakage, and “3” is called “image” thanks to IQ
imbalance [6].

By Criterion 71
I/Q Signals
 Consequently, calibration is necessary for improving IQ
imbalance and carrier leakage [13].

By Criterion 73
I/Q Signals
 As mentioned earlier, the XO signal from PMIC to transceiver
is rich in harmonics (i.e., 19.2 MHz*N, N is integer). Thus, IQ
traces should be away from XO signal. Otherwise, those
channels, whose frequencies correspond to 19.2 MHz*N,
may have EVM issue.

By Criterion 74
I/Q Signals
 Therefore, please reserve LC filter on IQ traces, as shown
below [13]:

By Criterion 75
I/Q Signals
 Nevertheless, the value of bypass capacitors should NOT be
too large. For example, the 150 pF frequency response is as
shown below:

By Criterion 76
I/Q Signals
 Since the self-resonant frequency of 150 pF is 847 MHz,
which exhibits LPF characteristic in baseband domain. The
IQ frequency range of various bandwidth is as shown below:
 Thus, IQ amplitude of the signal with wider bandwidth may
be attenuated, thereby causing power and EVM issues.

By Criterion 77
I/Q Signals
 Besides, avoid routing IQ lines near or directly under PMIC
SMPS (Switching Mode Power Supply) PCB areas since the
strong switching noise can couple magnetically and
electrically to IQ lines even though in the presence of GND
plane separation.

By Criterion 78
I/Q Signals
 As shown below, the case where the IQ traces are separated
from PMIC switching node by multiple ground layers, but it
is not recommended [30].

By Criterion 79
I/Q Signals
 As shown below, the IQ lines don’t overlap PMIC area, this is
recommended.

By Criterion 80
Lesson learnt_9: EVM issue due to cable loss
 The transmit EVM specification for LTE is as shown below:
 The measurement is as shown below:
It’s over the limit obviously.

By Criterion 81
 As mentioned earlier, both too high and too low feed-back
power make the output power too high finally thanks to
compensation mechanism. Moreover, cable loss setting is a
critical factor as well.
 While doing calibration, RGI 51 is expected to generate 22.5
dBm output power. Nevertheless, in the absence of cable
loss setting, RGI 51 generates merely 21.5 dBm output
power. Therefore, in order to achieve expected 22.5 dBm, the
RGI is increased to 52. But, with RGI 52, the actual output
power is 24 dBm in the presence of cable loss setting.

By Criterion 82

By Criterion 83
 Consequently, although the measured output power is 22.5
dBm on the display, the actual output power is larger than
the measured one, maybe 24 dBm. As mentioned earlier, the
AM-AM and AM-PM distortion increases as the output power
is increased, so does EVM. That’s why EVM failed.

By Criterion 84
 Usually, there will be some symptoms during calibration if
cable loss is NOT set well.

By Criterion 85
 As shown above, HDET value in mid channel is too high.
 During calibration, if the received power by CMW500 is
too low, it will force DUT to increase output power to
achieve target power, thereby causing too large output
power and feedback power.

By Criterion 86
 Thus, as shown below, if cable loss setting is lower than
actual cable loss, which makes the output power received
by CMW500 lower than expectation. And then, CMW500 will
force DUT to increase its output power

By Criterion 87
 Hence, increase additional 0.7 dB cable loss, and the
calibration passes.

By Criterion 88
 In short, during calibration
 Too high outputpower and feedback power => cable loss
 Too low output power and feedback power => cable loss
 So that’s why output power in low/high channels decreases
as the cable loss increases. And, of course, so does mid
channel.(So the calibration passes)

By Criterion 89
FBRX
 Have a look at the definition of EVM again, which includes
magnitude error.
 Thus, lay FRBX trace in inner layers for better protection.
Otherwise, magnitude error leads to poor EVM as well.

By Criterion 90
FBRX
 Why is there a LPF on FBRX path to suppress WiFi 5 GHz
signal ?

By Criterion 91
FBRX
 Due to LNA poor IP2, the IMD2(5380 MHz – 2690 MHz = 2690
MHz) may increase the noise floor, thereby aggravating HB
signal feedback SNR.

By Criterion 92
FBRX
 The WiFi 5 GHz signal may leak to FBRX path through ASDIV.

By Criterion 93
FBRX
 The DRX ANT may radiate WiFi 5 GHz signal to HB ANT.

By Criterion 94
FBRX
 Without LPF rejecting WIFI 5 GHz jammer on FBRX path,
LNA poor IP2 leads to high IMD2 product raising noise floor,
thereby aggravating HB signal EVM performance due to
magnitude error.

By Criterion 95
FBRX
 As shown below, there is NO DC_I/Q calibration data stored
in Gain State 3.

By Criterion 96
FBRX
 Again, we have to realize that FBRX is a LNA. So incorrect
gain mode leads to low SNR, thereby aggravating EVM
performance.
 Of course, DC block is necessary.

By Criterion 97
FBRX
 Is there something wrong for the block diagram?

By Criterion 98
FBRX
 From calculation, the SNR is merely 10 dB, NOT enough.

By Criterion 99
FBRX
 Therefore, daisy chaining these couplers for FBRX path is
forbidden due to insufficient SNR. The SNR should be at
least 30 dB, and which can be achieved by SPnT due to its
port-to-port isolation.

By Criterion 100
FBRX
 With SP4T, the SNR is acceptably 40 dB(> 30 dB).
SP4T port-to-port isolation

By Criterion 101
The influence of filter
 Filter contributes to EVM as well. As mentioned earlier, DPD
is a method to improve linearity. Nevertheless, during DPD,
the pre-distorted waveform will be truncated if the filter
bandwidth is not wide enough, thereby contributing to EVM
[5].

By Criterion 102
 Besides, the deviations in group delay cause signal
distortion [5].

By Criterion 103
 Usually, large group delay variation appears near the
transition region in frequency response, leading to distorted
waveform [5].

By Criterion 104
 Thus, with large deviations in group delay, the channels
near the transition region suffer from EVM issue more easily.
The total EVM of an LTE signal is as calculated below [5]:
 EVMi is the EVM measured across the individual RB. N is the
total number of RBs in the LTE signal.

By Criterion 105
 EVMi can be as calculated below:
 ∆α is the effective magnitude ripple across the individual RB
of the filter’s passband; ∆ø is the effective phase ripple
across the individual RB of the filter’s passband [5].
 Thus, both ∆α or ∆ø across the individual RB of the filter’s
passband have impact on the overall EVM performance.

By Criterion 106
 In addition to group delay ripple and bandwidth, temperature
stability is a crucial factor contributing to EVM as well. As
shown below, the frequency response may drift towards the
left side under high temperature.

By Criterion 107
 Thus, during calibration, perhaps the growing heat in PCB
makes the frequency response drift towards the left side,
thereby reinforcing the loss in high channel and the output
power is not as expected. At this time, as mentioned earlier,
the RGI is increased to compensate for the loss so as to
achieve expected power. This causes too high output power
in high channel under normal temperature, thereby
aggravating EVM [5].

By Criterion 108
Lesson learnt_10: EVM issue due to filter
 As shown below, by far the worst result was the 15 MHz
bandwidth case due to insufficient filter’s bandwidth (14.6
MHz), thereby truncating the waveform of some channels.

By Criterion 109
 Furthermore, except 15 MHz, it is apparently that narrower
bandwidth results in worse EVM. It is relevant for proportion.
For instance, with 1.4 MHz signal bandwidth, if three RBs are
contaminated by large group delay ripple near transition
region (e.g. low/high channel), it means that 50% RBs (3/6 =
50%) have poor EVM, thereby aggravating the overall EVM.
 Conversely, with 10 MHz signal bandwidth, even though 5
RBs are contaminated by large group delay ripple near
transition region, it means that merely 10% RBs (5/50 = 10%)
have poor EVM, which is not severe enough to the overall
EVM [5].

By Criterion 110
 Thus, what matters most is how much the proportion is, not
how many the contaminated RBs are.

By Criterion 111
Timing
 As mentioned earlier, TDD bands have worse linearity
performance than FDD bands. Additionally, for TDD PA, once
PA is on, amplitude must be flat during entire transmission.
Otherwise, any rise or droop contributes to AM/AM
distortion and degrades EVM [26]. So TX related timing
(PA_ON, ASM, etc.) is crucial [31].

By Criterion 112
Power
 Furthermore, thanks to dynamic mode operation, once PA is
on, the power supply generates huge transient current,
thereby aggravating voltage ripple[13].
 Any imperfection in power supply (e.g. IR drop, ripple, noise)
causes poor transmitter performance.
Thus, for TDD PA, proper decoupling method is necessary.

By Criterion 114
Power
 As for transceiver, pay attention to not only keep-out areas,
but also power supply. Use star-routing for power supply
pins rather than daisy-chain [30].
Power
Supply
Noisy pin Contaminated pins Noisy pin Clean pins
Power
Supply

By Criterion 115
Power
 As shown below, there are five power supply sources from
PMIC to transceiver, four of them (as marked green circle)
provide to multiple pins. It is necessary to make use of star-
routing for these pins.

By Criterion 116
Power
 Branch at capacitor (as marked black circle in previous
photo) only [30].

By Criterion 117
Power
 As shown below, the isolation effect depends on the branch
point position, which should be as close to power source as
possible to increase the impedance of the coupling path
between noisy pin and other pins, thereby making other pins
as clean as possible.
Noisy pin Clean pins
Power
Supply
Noisy pin Clean pins
Power
Supply

By Criterion 118
Lesson learnt_11 EVM issue due to LCM
 EVM fails while LCM is on, but passes while LCM is off. As
shown below, since VPH_PWR branches close to backlight
driver IC, thereby causing the impedance of the coupling
path low. Thus, transient current from backlight driver IC
leaks to MMPA easily. That’s why EVM passes while LCM off
due to the absence of transient current [13].

By Criterion 119
Lesson learnt_12 EVM issue due to WIFI
 EVM fails while LTE and WIFI operate simultaneously. As
shown below, there is no sufficient isolation between
cellular and WIFI XO traces. As mentioned earlier, XO signal
is rich in harmonics, so they interfere with each other in this
case [35].

By Criterion 120
Lesson learnt_13 EVM due to incorrect schematics
 As shown below, since TX_DAC1 IQ pin is connected to
GND, DC current leaks from GND to IQ pins that practically
function. Hence, carrier leakage causes EVM issue

By Criterion 121
Lesson learnt_13 EVM due to incorrect schematics
 Those IQ pins that don’t practically function should be
floating instead of shorting to GND. As shown below:

By Criterion 123
Introduction
 The IMD (Intermodulation) contributes to ACLR. Therefore,
the linearity of transmitter chain, especially PA, determines
the ACLR performance [6,27].

By Criterion 124
Introduction
 As mentioned earlier, SC-FDMA is used for LTE uplink
transmission to reduce PAR [1,5]. Nevertheless, LTE still has
higher PAR in uplink modulation than WCDMA. Thus, even
though in the presence of smaller output power (LTE:23
dBm, WCDMA:24 dBm), LTE still has worse ACLR
performance due to higher PAR. In other words, the linearity
requirement of LTE is more stringent than WCDMA.

By Criterion 125
Lesson learnt_14: WCDMA ACLR issue due to ICQ
 With the identical frequency range, LTE B5 and WCDMA B5
share the same transmitter path. Because the linearity
requirement of LTE is more stringent than WCDMA, there is
no need to tune PA output impedance to transform the load
impedance into the desired one. Finally, the issue was
solved by modifying ICQ.
 Thus, if LTE/WCDMA/GSM
share the same
TX path, you just need
to test LTE TX
performance during
matching tuning phase.

By Criterion 126
Synchronization for ET
 For ET, it is crucial that Vcc and the RF input signal are
aligned in time of the PA. Otherwise, there will be time delay
between RF input signal and Vcc, thereby aggravating
transmitter performance, such as ACLR and EVM [29].
Thus, proper time delay adjustment is necessary.

By Criterion 127
Lesson learnt_15: ACLR issue due to decoupling capacitor
 With ET technique, wider signal bandwidth aggravates ACLR
performance since PA with ET technique always closes to
saturation point, where nonlinearity becomes notable. And
wider LTE signal bandwidth brings more RBs, which lead to
higher PAPR and higher requirement of linearity.

By Criterion 128
 As shown below, ET DC-DC converter provides power
supply to PA (i.e., Pin29, VSW), and C5513 (as marked blue
rectangle) should be 470 pF.

By Criterion 129
 Nevertheless, thanks to incorrect value (470 pF -> 47 nF),
and the signal waveform is as shown below:

By Criterion 130
 In terms of time domain, high PAPR means high variation in
envelope. Since bandwidth of 20 MHz has higher PAPR than
5 MHz, which means the aforementioned synchronization is
more crucial to 20 MHz than 5 MHz bandwidth.
 Hence, too large decoupling capacitor value slows down the
speed of Vcc and reinforces time delay between Vcc and RF
signal, especially in 20 MHz bandwidth, and ACLR fails.

By Criterion 131
 Definitely, now that the root cause is synchronization. As
long as we use conventional fixed power supply instead of
ET, the issue is gonna be solved as well.

By Criterion 132
GSM PA and 3G/4G PA for ET configuration

By Criterion 133
 As shown above, now that path1(red path) supports both
APT/ET, why not make GSM PA using APT and 3G/4G PA
using ET share the same Vcc?
 Take Qualcomm QET4100 for example, its load capacitance
should NOT be more than 900 pF.

By Criterion 134
 In general, the total internal shunt capacitance of Vcc of
3G/4G PA is about 100~300pF.
 Thus, even though two or three 3G/4G PAs share the same
Vcc, their total internal shunt capacitance is still less than
ET DC-DC converter load capacitance limit.

By Criterion 135
 Compared to 3G/4G PA, GSM PA has large total internal
shunt capacitance, even as large as 2 ~ 3 nF.
 So making 3G/4G PAs and GSM PA share the same Vcc
makes the total internal shunt capacitance larger than the
limit by a long shot. Doing this brings the aforementioned
synchronization issue definitely.

By Criterion 136
Memory Effect
 Occasionally, the ACLR is asymmetric, which is related to
memory effects in PA. Memory effects are changes in a PA’s
nonlinearity resulting from the previous history of the input
signal [32]. Self-heating has already been proven to be one
of the key sources to memory effect in PA. In addition, the
memory effect depends on signal bandwidth as well [33].
Therefore, asymmetric ACLR phenomena often occurs in
maximum output power, especially wider bandwidth.

By Criterion 137
Memory Effect
 The solution to this issue is to tune PA load impedance [34].

By Criterion 138
Non 50 Ohm impedance of connector
 Tune the impedance from antenna port to duplexer first to
shrink the circle, then tune the impedance from PA to
duplexer to determine the circle location in Smith Chart

By Criterion 139
 Nevertheless, in this case, for connector signal pad, to see
L9 as GND proves severe mismatch due to merely 14 Ohm
[13].

By Criterion 140
 Therefore, in this case, it is very difficult to pull load
impedance to 50 Ohm for the common path [13]. Thus, for
PA, ASM. and connector etc., metal under their signal pads
should be cut out to retain 50 Ohm, if necessary. In this case,
L9, L8, and L7 should be cut out to retain 50 Ohm (GND is
L6).

By Criterion 141
Harmonics of LO
 Ideally, the baseband signal is mixed with an up-converter
(LO) to obtain the (LO + BB) component at transceiver
output. Nevertheless, practically, the LO often generates
square-wave signal which is rich in harmonics. Thus, there
will be (LO ± BB) and (3LO ± BB) at mixer output.

By Criterion 142
Harmonics of LO
 Thus, if not properly filtered before PA, (LO ± BB)
components due to PA nonlinearity appears near RF signal,
thereby aggravating ACLR [31].

By Criterion 143
Lesson learnt_16: ACLR issue due to charging
 ACLR only fails while charging function enables. For a 5V
DCP (Dedicated Charging Port) plug-in, the charging noise
is set to 600 kHz, which leaks from PMIC through PA DC-DC
converter to PA. Hence, the IMD2 components near RF
signal aggravates ACLR.

By Criterion 144
Lesson learnt_16: ACLR issue due to charging
 So the filter is necessary on the path between PMIC to ET
DC-DC converter, or the direct path between PMIC and PA.

By Criterion 145
Lesson learnt_17: ACLR issue due to shielding can
 As shown below, if someone presses the shielding can with
finger, the ACLR performance is good, and vice versa.

By Criterion 146
 As mentioned earlier, the PA couples its strong RF energy
onto the shielding can. An appreciable fraction of the PA
output couples to the PA input by means of reflection,
thereby driving PA to higher output power level (25 dBm)
and aggravating ACLR. To press the shielding can with
finger is able to reinforce the grounding, thereby making all
the fraction of PA output signal flow to main ground directly.
By doing this, no reflection, no issue.

By Criterion 147
 Furthermore, if the reflected TX signal leaks to PA Vcc, as
mentioned earlier, any imperfection in power supply such as
noise, TX performance will become poor.

By Criterion 148
PA input matching
 Additionally, PA input is the load impedance of DA (Driver
Amplifier) as well. In other words, non-50 Ohm impedance
degrades DA’s linearity and aggravates ACLR at PA input.
Poor ACLR at PA input causes worse ACLR at PA output.

Performance requirement and lessons learnt of LTE terminal---transmitter part

More Related Content

What's hot (20)

Similar to Performance requirement and lessons learnt of LTE terminal---transmitter part (20)

More from criterion123 (8)

Recently uploaded (20)

Performance requirement and lessons learnt of LTE terminal---transmitter part