Matt Ozalas / Keysight

Tom Winslow / Macom

1
• Learn Techniques to identify and solve stability problems in the design phase
• Learn a new way to simplify stability analysis using the WS-Probe
• See an example of the techniques applied to a real design

Avoid seeing this in the lab

(unless you want to design an oscillator!)

2
 W H Y I S T H I S I M P O R T AN T ?

• Current state of the wireless communications industry:

• Systems are getting more complex, higher in frequency … no matter the application!
• Circuits are becoming highly integrated and compressed, using devices with higher gain.

• Instability fundamentally derives from a combination of gain and feedback.

• In the past, designers might add loss or adjust bypassing in the lab to stabilize circuits

• Today, high frequency designers need to master stability by truly understanding the root causes
early in the development process, and incorporating this knowledge into all aspects of design

3
 O N LY A P P L I E S T O S T A B L E U N L O A D E D N E T W O R K S !

1 − 𝑆11 2 − 𝑆11 2 + ∆ 2
• Premise: A device is unconditionally stable if: 𝐾 = >1 ∆= 𝑆11 𝑆22 − 𝑆12 𝑆21 < 1
2 𝑆12 𝑆21

• Platzker, Struble. “Instabilities Diagnosis and the Role of K in Microwave Circuits”, IEEE 1993:
“Classic” Ring Oscillator Example shows K stable, but more rigorous analysis (NDF) shows the circuit is unstable

4
General / Mathematical Approaches

Eigenvalues
Normalized Null
Active / Passive Bilateral
Determinant Injection
Imbedding
Unilateral
Pole-Zero Auxiliary Direct Loop
Decomposition Generator Calculation Analysis

Driving Return
Impedance Difference

Stability Factor BELL LABS,

(K / Mu) 1920’s-40’s
Bode, Nyquist, Black

5
General / Mathematical Approaches Solve Voltage /
Turn off True Return
matrix Current Stimuli
Simulation techniques sources Ratio
Circulators
Eigenvalues Osctest
Normalized Null
Active / Passive Bilateral
STAN Determinant Injection
SS Inject Imbedding
Unilateral
Pole-Zero Auxiliary Direct
Bifurcation Loop
Decomposition Generator Calculation
Analysis Analysis
Break the loop
Driving Transient Return Inject signals
Impedance Analysis Difference

Stability Factor BELL LABS,

(K / Mu) Direct compute
S-probe 1920’s-40’s for trivial devices
Apply Bode, Nyquist, Black
functions
Compute: Determinant, Gain / Phase Margin, Origin …Or, push simulator into multiple solution 6
encirclements, Zero crossings, negative resistance… space / directly stimulate oscillation condition
 A P P LY M U L T I P L E S T A B I L I T Y T E C H N I Q U E S W I T H O N E S I M U L AT I O N

Step 1: Add Probes

Step 2:Apply equations to derive common figures of merit

7
Agenda:
• Introduction
• Background: Review of Feedback Systems
• Assumptions can make the results difficult to interpret

• Fundamental Concepts in Stability Analysis

• Return Ratio and Driving Point Admittance

• Unifying simulation approaches using in-situ probing

• Introducing the “WS_Probe”
• Circuit Design Example
• MMIC Power Amplifier
• Closing and Q&A

8
9
 a*[In+ f*Out]
In Ʃ +
In+f*Out
a Out 𝑂𝑢𝑡 = 𝑎 𝐼𝑛 + 𝑓 ∗ 𝑂𝑢𝑡
𝑂𝑢𝑡 = 𝑎 ∗ 𝐼𝑛 + 𝑎𝑓 ∗ 𝑂𝑢𝑡
𝑂𝑢𝑡 ∗ 1 − 𝑎𝑓 = 𝐼𝑛 ∗ 𝑎𝑓
f *Out
f 𝑂𝑢𝑡 𝑎
𝐴𝑐𝑙 = = Loop Gain
𝐼𝑛 1 − 𝑎𝑓

∞ 𝑎
𝐴𝑐𝑙 =
1 − 𝑎𝑓0
A closed loop network is unstable
when loop gain (af) is equal to 1
1 − 𝑎𝑓 = 0 ⇒ 𝑎𝑓 = 1

10
• The Loop Transfer function is really an S-domain equation:
𝐴 𝑠 𝑠−𝑧1 𝑠−𝑧2 …
𝐴𝑐𝑙(𝑠) = =
1+𝐴 𝑠 𝐹(𝑠) 𝑠−𝑝1 𝑠−𝑝2 …

• The poles (denominator) determine the form of the time domain system function
• System poles in the S-domain transform into exponential functions in the time domain.
𝐾 𝐴1 𝐴2
= + ֞ 𝐴1 𝑒 𝑝1𝑡 + 𝐴2𝑒 𝑝2𝑡
𝑠 − 𝑝1 𝑠 − 𝑝2 (𝑠 − 𝑝1 ) (𝑠 − 𝑝2 ) Key Takeaway: Avoid the
Dreaded “Right Half Plane”

Im{s} Poles having negative real parts result

Im{s}
Poles having positive real parts result
in exponentially decaying functions in exponentially growing functions

Re{s} Re{s}

𝑠 − (−𝑝𝑖 ) = 0 → 𝑠 = −𝑝𝑖 𝑠 − 𝑝𝑖 = 0 → 𝑠 = +𝑝𝑖

𝐴1 𝑒−𝑝1𝑡 + 𝐴2 𝑒−𝑝2𝑡 𝐴1 𝑒𝑝1 𝑡 + 𝐴2 𝑒𝑝2𝑡
11
Apply Cauchy’s Principle To Find denominator zeros
𝑎(𝑠)
Clockwise origin encirclements= 𝐻(𝑠) =
Zeros - Poles 1 − 𝑎(𝑠)𝑓(𝑠) 1-af=0
Sweep S
around ANY j∞
Sweep s over
closed contour
unstable RHP
j0
Zeros=4
+∞
Poles=2

-j∞
H(s) af
Observe Count CW
1-af origin
system
output encirclements

1 encirclement
2 more zeros than poles,
2 clockwise encirclements
means 1 RHP zero 12
Feedback →
Passive RLC
Network

Amplifier →
Transistor Model

13
 Stable, but close!

Result is a loop gain “S-parameter”

14
 3

a
1
ZL≠ZT
2
In Ʃ + Out
a
ZT

15
MIDDLEBROOK, 1975

Measurement technique (HP 302A)… adapted for simulation as follows:

Apply test voltage source Apply test
current source

1. Compute Voltage Loop Gain 2. Compute Current Loop Gain Unstable

3. Combine for total loop gain

16
HURS T, 1 9 9 1

y12f*vo 𝐼𝑓𝑏
Feedback y22f (y12a+ y12f )vo 𝑓≜
y11f 𝑉𝑜
y21f*vi
f(s)

(y21a+ y21f )vi

Amplifier y21a*vi y11a y22a
y11a y22a 𝑉𝑜
a(s) y12a*vo + + 𝑎≜
y22f 𝐼𝑎
y11f

𝐼𝑓𝑏 (𝑌21𝑎 +𝑌21𝑓)(𝑌12𝑎 +𝑌12𝑓)

𝑎𝑓 = 𝐼𝑎
= (𝑌
11𝑎 +𝑌11𝑓 )(𝑌22𝑎 +𝑌22𝑓 )
Stable
Bilateral Loop Gain is Derived from Y-
parameters of the amplifier / feedback blocks 17
18
WHICH ONE IS CORRECT ?

K-factor: Unstable Osctest: Stable Null injection: Unstable Bilateral model: Stable

19
• Bode, 1943: “Network Analysis and Feedback Amplifier Design”.

Driving Return Ratio

Impedance

BELL LABS,
1920’s-40’s
Bode, Nyquist, Black

20
• Return Ratio and Return Difference
• A system can be described by a transfer function H(s)
• Transconductance elements are potential sources of instability
• Analysis can be performed at the element to characterize:
• F=Return Difference (Feedback) ; T=Return Ratio (Loop Gain)
• Return difference is the ratio of the system determinant to the
same determinant with the gain element removed.

• Driving Point Impedance or Admittance

• A system can also be described by a set of nodal equations
• Solutions to the equations represent the steady state
• Driving point impedance is the ratio of the system determinant
to the same determinant where the row and column containing
the node under analysis is removed

21
• Return Ratio and Return Difference
• A system can be described by a transfer function H(s)
• Transconductance elements are potential sources of instability
• Analysis can be performed at the element to characterize:
• F=Return Difference (Feedback) ; T=Return Ratio (Loop Gain)
• Return difference is the ratio of the system determinant to the
same determinant with the gain element removed.

• Driving Point Impedance or Admittance

• A system can also be described by a set of nodal equations
• Solutions to the equations represent the steady state
• Driving point impedance is the ratio of the system determinant
to the same determinant where the row and column containing
the node under analysis is removed

22
Sweep Source gm Toggle
(on / off)
𝑌 = 𝑠𝑡𝑜𝑦(𝑆, 1𝑒12)
Hi-Z Terminations (DC Blocked if needed)

det(𝑌)
𝐹= =1+𝑇
det(𝑌0 )
Return Difference and Return Ratio are derived from the Y-matrix
determinant, with the internal transconductance toggled on and off 23
 ∆ (determinant)
𝑦𝑑𝑝(𝑘) =
∆𝑘𝑘 (determinant, with kth row and column removed)

𝑦11 𝑦12
1 2
𝑦21 𝑦22

Removing “node 1”:

Zero (first row and column):
crossing 𝑦11 𝑦12
𝑦
𝑦21 𝑦22 → 22
Kurokawa’s 𝑦11 𝑦22 − 𝑦21 𝑦12
condition for 𝑦𝑑𝑝(1) =
oscillation
𝑦22
Driving Point Admittance is derived from the Y-matrix,
normalized with the node under analysis removed
24
S T R U B L E A N D P L AT Z K E R , 1 9 9 3

Δ Accounts for
Δ𝑁𝐷𝐹 = = 𝑅𝑅1 + 1 𝑅𝑅2 + 1 𝑅𝑅3 + 1 … (𝑅𝑅𝑁 + 1) multiple sources
Δ0𝑁
Built-in ADS determinant function,
improved to support NDF:
Encirclements function in ADS

Sweep toggle
<off/on>

8 High Z
“Probes”
Internal
source toggle
Matrix order is expanded based on the
Transistor model with access to number of active sources in the design 25
internal 4 port transconductance
 𝐼𝐴𝐺 1
𝑌𝐴𝐺 = =
𝑉𝐴𝐺 𝐻0

REAL PART <0

IMAG PART = 0

Y_AG can be monitored for Kurokawa’s condition directly:

Re{Y}≤0; Im{Y}=0; Im{Y} increasing with frequency
26
T I AN , 2 0 0 1

Vin=Vout,
Network reduces to 1 port
𝑦11 𝑦12 𝐼
[𝑌] = 𝑦 𝑦22 vo
𝑌1𝑃 = 𝑌1𝑃 = = y11+y22 + y21+y12
21
𝑉
[Y] y11
+ 𝑌1𝑃 = y11+y22
(y21+y12)vo 0
y22
Feedback
𝑌1𝑃 y11+y22 + y21+y12
𝐹= =
Amplifier 𝑌1𝑃 0 y11+y22
vo
y21+y12
y12*vo = 1+
y11 y22 y21+y12 y11+y22
y21*vo 𝐿𝐺 = −
y11+y22 This is the return
difference

Despite the name, “True Return Ratio” is really

another derivation of Loop Gain
27
In our example, “True” Return Ratio does not match the internal calculation…

Bode Definition RR
(Despite this, the prediction of instability transition is still accurate)

Tian True RR
Why the difference? Transistor Model is not the same form as the derivation assumed:
To make results agree perfectly, one must simplify the transistor as follows:

Bode Definition RR
Tian True RR
28
 Driving Admittance combines two stability measures: Loop Gain, and Port Impedance

IAG [Y] 𝐼𝐴𝐺 = 𝑉𝐴𝐺 ∗ (𝑦11 + 𝑦12 + 𝑦21 + 𝑦22 )

y11 y12 y21 y22 𝑌𝐴𝐺 = 𝑦11 + 𝑦12 + 𝑦21 + 𝑦22

VAG 𝑦21 + 𝑦12
𝐿𝐺𝑇𝑅𝑅 = − Bilateral “True Return Ratio”
𝑦11 + 𝑦22
1
𝑌𝑜𝑝 = ≝ 𝑦11 + 𝑦22 “Open Loop” Impedance
𝑍𝑜𝑝 (see work by Ochoa)

𝑌𝐴𝐺 = 𝑌𝑜𝑝 (1 − 𝐿𝐺𝑇𝑅𝑅 )

29
 Battery
Design
Matching
Areas of
high out of
band loop
gain
f0
PCB

30
• Return Difference → Normalized Determinant Function
• Global, Rigorous Stability Technique…but, need access to nodes, and does not specify cause.

• Driving Point Admittance → Auxiliary Generator

• Local, Rigorous Stability Technique, but requires analysis on a node-by-node basis.

• Loop Gain is a useful design tool when combined with one of the above
• But, doing so requires multiple separate simulation benches.

Driving Point Return

Admittance Difference

BELL LABS,
1920’s-40’s
Bode, Nyquist, Black
31
32
 Hurst Loop Gain
External NDF Driving Point Admittance
Osctest Middlebrook/TRR Loop Gain

Bi-directional Impedance Probing (S-Probe) has the potential to simplify, but ….

33
 0.5pF 4pF 8pF
Low Frequency
4pF 8pF
1nH 0.5nH
S 50Ω 20Ω
Z1 Z2 Series R =50 Z1 Z2 Series R =20
50Ω 20Ω Series C=4p Series C=8p

Z1 and Z2 Series R Z1 and Z2 Series C High Frequency

0.5pF
S-probe gives negative
asymptotic capacitance 4pF 8pF
at high frequency,
should sum to 0.42 pF

S-probe gives negative real

impedance at high frequency,
should be zero Series R =0 Z1 Z2
Series C=Z1+Z2 =0.42p
34
 0.5pF 4pF 8pF
Low Frequency
4pF 8pF
1nH 0.5nH
ws 50Ω 20Ω
Z1 Z2 Series R =50 Z1 Z2 Series R =20
50Ω 20Ω Series C=4p Series C=8p

Z1 and Z2 Series R Z1 and Z2 Series C High Frequency

0.5pF

4pF 8pF
WS-probe outputs sum to
0.42 pF at high frequency

WS-probe gives zero real

resistance at high frequency Series R =0 Z1 Z2
Series C=Z1+Z2 =0.42p
35
 Red = Derived from single unified simulation with WSP
Blue = Earlier individual simulation result

Single Testbench, probes applied to amplifier

WS-Probes added to input and

output of amplifier model…

36
 Red = Derived from single unified simulation with WSP
Blue = Earlier individual simulation result

WSP Simulation = Osctest Simulation

WSP can virtually apply an Osctest termination

value after simulation, and reverse direction

37
 Red = Derived from single unified simulation with WSP
Blue = Earlier individual simulation result

==

WSP Simulation = Middlebrook LG Simulation

WSP can compute Loop Gain with just one simulation

(vs. two for the Middlebrook AC example)

38
 Red = Derived from single unified simulation with WSP
Blue = Earlier individual simulation result

==

WSP Simulation = Hurst LG Simulation

WSP can use probe pairs to denote “blocks” of

impedance for the Hurst Loop Gain Calculation

39
 Red = Derived from single unified simulation with WSP
Blue = Earlier individual simulation result

==

WSP Simulation = TRR Simulation

WSP can compute TRR with just one

simulation (vs. two for the AC example)

40
 Red = Derived from single unified simulation with WSP
Blue = Earlier individual simulation result

==

WSP Simulation = NDF Simulation

WSP can compile Y-matricies for any number of

probes acting as virtual terminations, for NDF

41
 Red = Derived from single unified simulation with WSP
Blue = Earlier individual simulation result

==

WSP Simulation = DPA Simulation

WSP can analyze any number of nodes for DPA with

just one sim (vs.1 sim /node in Aux Gen method)

42
• Here is the traditional method for taking one condition (Osctest) and performing a large signal
analysis on the circuit, rather than a small signal analysis. It is an entirely different methodology!

Monitor Return
Inject Small Signal

43
• Check the “Perform Stability Analysis” Box in the Harmonic Balance Controller…

44
 OR

WSP provides large signal results in the

same format as small signal results
45
• Ws-probe allows simultaneous analysis of multiple figures of merit with just one
simulation, as opposed to multiple simulations and benches.

• WS-Probes can also be used to derive impedance and admittances

• Can view bi-directional and open loop impedance and admittances (useful for design)
• Probes can be grouped together into “blocks” of two to return block level impedance, admittance
• Probes can be grouped into larger sets to compute admittance matrices, useful for NDF

46
47
 Power Supply Bypassing

Winslow Probes
Power Supply Bypassing

Stabilizing R

48
49
• EM / Circuit Excitation using Pathwave RF Pro is
a powerful stability diagnostic tool

• At low frequencies (100 MHz), EM-circuit excitation

reveals the loop through the bias network

• At high frequencies (3.4 GHz), EM-circuit excitation

reveals the loop through the matching network
ground

Bias line
coupling
Matching
ground
coupling

50
51
• The Winslow Probe in Pathwave ADS is a new way to simplify stability analysis,
allowing you to reduce the number of simulation setups and testbenches needed
to understand stability in high frequency circuits

• Download Example Workspace at the link provided in the resources tab.

52
• W. Struble and A. Platzker. “Instabilities Diagnosis and the Role of K in Microwave Circuits”, IEEE 1993:
• H. Bode, “Network Analysis and Feedback Amplifier Design”, D. Van Nostrand Co. Inc., New York, 1945.
• W. Hewlett, et al: “Some Applications of Negative Feedback with Particular Reference to Laboratory Equipment”,
Proceedings of the I.R.E, October 1939.
• RD Middlebrook: “Measurement of Loop Gain in Feedback Systems”, Int.J. Electronics, 1975.
• P.J. Hurst, “Exact Simulation of Feedback Circuit Paramters”. IEEE Trans. On Circuits and Systems, 1991.
• M. Tian, “Striving for Small-Signal Stability”, IEEE Circuits and Devices, 2001.
• W. Struble and A. Platzker, “A Rigorous Yet Simple Method for Determining Stabiltiy of Linear N-port Networks”, 15th Annual
GaAs IC Symposium Digest, 1993.
• A. Suarez, R. Quere, et. al, “Large Signal Design of Broadband Monolithic Microwave Frequency Dividers”, IEEE MTT-S
Digest, 1992.
• J.M. Collantes et al. “Expanding the Capabilities of Pole-Zero Identification Techniques for Stability Analysis”, IEEE
Microwave Theory and Techniques International Symposium, June 2009, Boston.
• A. Ochoa, “Loop Gain in Feedback Circuits: A Unified Theory Using Driving Point Impedance”, IEEE 2013.
• T. Winslow, “A Novel CAD Probe for Bidirectional Impedance and Stability Analysis “, IEEE 2018.
53
54