EE 489

1
EE 489 Telecommunication Systems
Engineering University of Alberta Dept. of
Electrical and Computer Engineering Switching
Systems Wayne Grover slides based on material
developed by W. Grover for EE589, (1998-2002)
set in powerpoint with a few additions by J.
Doucette 2002
2
Switching

• Circuit Switching
• A path is established between the caller and
  destination.
• Real-time connection formed.
• Example PSTN
• Message Switching
• Also called store and forward.
• A message is first stored in a buffer and then
  sent on in its entirety step by step as resources
  become available.
• No real-time connection (i.e. connectionless).
• Example E-mail
• Packet Switching
• A message is broken down into parts and each part
  is sent separately (possibly via different
  routes).
• Example Internet UDP protocol

3
Separating Circuits

• Four technologies for separating circuits
• Space, RF frequency, time, optical wavelength
• We want to logically connect circuits coming into
  a switch with circuits at the output.
• Example space division equivalent
  interconnection pattern

4
Space Division Switching

• Connecting two channels that are separated in
  space.
• Can be mechanical and/or electronic.
• Several problems
• Slow
• Bulky with lots of interconnect wiring
• Subject to cross-talk

5
Strowger Switching

• Patented 12/March/1889 and in some places still
  in use today.
• First widely-used automatic exchange system.
• A wiper assembly (contact arm) moves across a
  fixed set of switch contacts (contact bank).
• Each contact is connected to an outgoing channel.

6
Strowger Switching (2)

• Several uni-selectors can be graded together so
  multiple incoming circuits can connect to
  multiple outgoing circuits.

• Or two uni-selectors can be wired back-to-back
  (line-finders).
• 1st uni-selector chooses the incoming circuit,
  the 2nd chooses the outgoing circuit.

Unless there is heavy traffic, it is inefficient
and uneconomical to provide each incoming circuit
with a uni-selector.
Line-finders can be graded together as well to
form large switches.
7
Strowger Switching (3)

• In general, multiple uni-selectors, line-finders,
  and two-motion selectors (movable in two planes)
  can be connected in series.
• These switches respond to dialled digits,
  automatically switching an incoming circuit to
  the correct outgoing trunk.
• Step-by-step switching will respond to each digit
  individually.

8
Crossbar Switching

• Crossbar switching became popular in the 1940s
  and is still used in some places today.
• Uses a simple rectangular matrix.
• Actuators are operated at incoming circuits and
  outgoing circuits to make metallic contact and
  form the desired connection.

9
Time Division Switching

• In digital TDM systems (e.g. DS1), channels are
  divided by time slot, but switching is still
  possible.
• Switching is by a time-slot interchanger (TSI)
  and is accomplished by rearranging the order in
  which data is read out of the buffer.
• Incoming data enters a speech store while the
  outgoing channels indicate to the speech address
  memory (SAM) which incoming timeslot it is
  assigned to.
• During each time-slot, the outgoing circuit reads
  the speech store slot corresponding to the SAM.

10
Optical Switching

• One wavelength (or colour) can be turned into
  another.
• Called wavelength conversion or translation.
• Important in reducing blocking due to wavelength
  contention in routing and wavelength assignment
  (RWA) problem.
• Optoelectronic conversion consists of optical
  receiver, conversion to electronic signal (O/E),
  and then transmitter generates optical signal at
  the desired new wavelength (E/O).

11
Optical Switching (2)

• One (or several wavelengths) are switched from
  one fibre into another.
• Can use splitters and tunable filters, or
• More recently - Micro-Electro-Mechanical Switches
  (MEMS)
• On the scale of a human hair (100 microns)

Source Lucent Technologies - Bell Labs
Web-site http//www.bell-labs.com/news/1999/novem
ber/10/1.html http//www.bell-labs.com/org/physica
lsciences/timeline/1999_mems_expansion.html
12
Switching Network Design

• Several Points to Consider
• Blocking versus non-blocking switches
• Number of cross-points (i.e. size of the switch)
• Reliability
• Overload
• Growth
• Cost and technology
• Trunk Switch (aka traffic switch)
• One-to-one connection.
• One specific inlet must connect to one specific
  outlet.
• Access Switch
• One-to-any connection.
• One specific inlet must connect to any free
  outlet.

13
Multi-Stage Switch Fabrics

• Consider a switch with a 100 x 100 interconnect
  function.

Need 10 000 cross-points.
E.g.
If bi-directional transmission, then connection
from A to B is equivalent to a connection from B
to A (and connection from A to A is meaningless).
Need 4950 cross-points.
E.g.
14
Multi-Stage Switch Fabrics (2)

• Full matrix and even folded matrix switches may
  be inefficient since they scale as O(n2).
• A third method of achieving a 100 x 100
  interconnect function is by splitting the switch
  into two stages using smaller square matrices as
  building blocks.
• Then to form a connection, two xpts are operated,
  one in each stage, but
• (i) fewer xpts needed in total
• (ii) we may have introduced some blocking
  probability
• Example 100 x 100 in 2 stages

How many xpts?
Each block is 10 x 10 100 xpts.
Each stage is 10 blocks 1000 xpts.
Whole switch has 2 stages 2000 xpts.
15
Multi-Stage Switch Fabrics (3)

• How does it work?
• Divide the 100 inlets into groups of 10.
• 1st outlet of each Stage 1 block is connected to
  an inlet of the 1st Stage 2 block.
• 2nd outlet of each Stage 1 block is connected to
  an inlet of the 2nd Stage 2 block.
• 3rd outlet of each Stage 1 block is connected to
  an inlet of the 3rd Stage 2 block
• ith outlet of each Stage 1 block is connected to
  an inlet of the ith Stage 2 block.

16
Multi-Stage Switch Fabrics (4)
Example 16x16 2-stage switch using 4x4
non-blocking full matrices
Using this example, we can see every path through
the switch to connect any inlet in the 1st stage
to any outlet in the 2nd stage.
Again, notice the connection pattern The jth
outlet of the kth Stage 1 block is connected to
the kth inlet of the jth Stage 2 block.
Using any size of n x n blocks, you can make an
n2 x n2 2-stage switch.
We can also add a 3rd stage to the switch to get
an n3 x n3 3-stage switch How?
17
Multi-Stage Switch Fabrics (5)
Adding a 3rd stage to a 2-stage switch
Treat the original n2 x n2 2-stage switch as its
own block, attach it to n2 new blocks of n x n
and use the same connection pattern The jth
outlet of the kth Stage 1 block is connected to
the kth inlet of the jth Stage 2 block. Then copy
the original n2 x n2 2-stage switch n times and
repeat.
How many xpts? 27 x 27 3-stage switch 243 27 x
27 1-stage full matrix 729
18
Multi-Stage Switch Fabrics (6)
How many xpts? 1000 x 1000 3-stage switch 30
000 1000 x 1000 1-stage full matrix 1 million
Connection pattern used is called distribution,
and in general Stage n - Module k - Outlet j
connects to Stage n1 - Module j - Inlet k
Example Stage 2 - Module 1 - Outlet 91 connects
to Stage 3 - Module 91 - Inlet 1
19
Link Blocking

• Because of the single link between each module
  and the modules in the next stage, theres a
  possibility of blocking.
• Consider an inlet in the 1st block of stage 1
  connected to an outlet in the 3rd block of stage
  2.
• Now what happens if we want to connect another
  inlet the 1st block of stage 1 to another outlet
  of the 3rd block of stage 2?

A problem arises because there is only a single
route available through a switch with only
distribution-type of stages.
? Even though the entire switch is made up of
non-blocking square matrices, we can still
encounter blocking.
20
Estimating Blocking

• Distribution stages increase the overall
  inlet/outlet size of the switch but introduce
  increasing probability of blocking.
• there is only a single path between any specific
  1st stage inlet and any specific final stage
  outlet.
• Mechanism of blockage is when an inter-stage link
  on required path is in use.
• The greater the number of links in the path, the
  greater the probability that one of them is in
  use.
• Therefore, the more distribution stages we have,
  the greater the probability of blocking (but the
  larger the total switch size is).

21
Estimating Blocking (2)

• For a Pure Distribution switch
• Say we have a Erlangs of traffic on an inlet,
  then the proportion of time it is used is also a,
  and
• assume that all connections are random, and so
  the probability of any one link being occupied is
  also a in any stage (if we use square blocks),
  so
• Probability of any specific link being free is 1-
  a.
• But we need all links in the path to be free so
  probability that the path is available is (1-
  a)k-1.

22
Mixing Stages

• Weve seen that we can add distribution stages to
  increase the switch size nk x nk (where n is the
  size of each square matrix block, and k is the
  number of distribution stages), but
• We need a way of reducing blocking.
• The solution is to add a mixing stage (also
  called collection stage) that keeps the overall
  switch size the same (in terms of nk inlets and
  outlets), but can reduce blocking by adding
  multiple paths through the switch.

Distribution
Distribution
Mixing
23
Mixing Stages (2)
Adding a 3rd distribution stage to a 2-stage
switch
Adding a mixing stage to a 2-stage switch
Connection pattern is the same as for
distribution Stage n - Module k - Outlet j
connects to Stage n1 - Module j - Inlet k
The difference is that we dont replicate the
2-stage switch n times.
24
Mixing Stages (3)

• By how much does a mixing stage reduce blocking?
• Adding a mixing stage will provide n alternate
  paths through the switch.

Example (n 3)
Recall that probability of blocking of each path
is
But for blocking to occur, we must have all n
paths blocked
25
Call Packing

• Analyze how blocking in a network occurs
• There are generally free links in each stage.
• Problem is that they are mismatched from stage to
  stage.
• For instance

Even though there are free links throughout the
switch, there is a conflict for specific links
required for the brown connection.
26
Call Packing (2)

• Call packing is a strategy of organizing new
  calls so that they use free links corresponding
  to other busy links in the next stage if possible.

By appropriately packing the other connections,
the brown connection can now find an available
path.
27
Clos Non-Blocking Switches

• Consider the call blocking mechanism

The brown connection cant find a path through
the switch.

• Is there a way of designing the switch with
  appropriately sized modules and stages so that
  its impossible for there to be blocking, even if
  without call packing?

28
Clos Non-Blocking Switches (2)

• Consider the worst possible case
• Connect from an inlet in a first stage module
  (with n inlets) where n-1 of its outlets are
  already in use to an outlet in a final stage
  module where (with n outlets) where n-1 inlets
  are already in use, and none of the busy links
  are matched.

Need one extra module to connect through.
29
Clos Non-Blocking Switches (3)
? For a guarantee of a free path through the
switch, we need (n-1)(n-1)1 2n-1 modules in
the 2nd stage, and
each 1st stage module needs 2n-1 outlets,
and each 2nd stage module needs N/n outlets and
inlets, and each 3rd stage module needs 2n-1
inlets.
30
Clos Non-Blocking Switches (4)

• A (non-blocking) Clos switch will have the
  following structure

Can also show that to minimize number of xpts
31
To Minimize Number of Cross-Points
To minimize the number of xpts
32
5-, 7-, 9-Stage Clos Switches

• Clos switches can be nested together.
• Middle stage modules themselves are
  appropriately-sized 3-stage Clos switches.

• Why would we want to do this?
• Each module is non-blocking (whether full matrix
  or Clos network).
• If we use Clos networks, we have fewer xpts.

33
Digital Switching

• Time Slot Interchanger (TSI).
• A TSI is a time switch.
• Switches one time slot channel in a single
  physical input to another time slot channel on a
  single physical output.
• Functionally equivalent to an n x n space-divided
  switch where n is the number of time slots per
  frame.

• Time multiplexed space switch (TMSS)
• A space switch (multiple physical inputs and
  outputs) that is potentially reconfigured
  entirely in every time slot of each frame.
• Data is switched such that for each time slot,
  specific inlets are connected (switched) to
  specific outlets.
• Data does not switch timeslots.

34
Time Slot Interchanger

• In a TSI, one time slot is switched to another.
• Performed through use of two memory stores
• Speech store is RAM with capacity to store one
  full frame of data.
• For DS1 (1.544 Mbps) with 24 channels of 8 bits,
  the speech store is 24 bytes long.
• For E1 (2.048 Mbps) with 32 channels of 8 bits,
  the speech store is 32 bytes long.
• Speech address memory (SAM) or Time Switch
  Connection Store is RAM with capacity to store a
  word for each time slot, each word being a
  number identifying a specific time slot.
• For DS1, the SAM has capacity to store 24 words
  of 5 bits per word (need 5 bits to store a number
  between 1 and 24) for a total of 24x5 bits.
• For E1, the SAM has capacity to store 32 words of
  5 bits per word (need 5 bits to store a number
  between 1 and 32) for a total of 32x5 bits.

35
Time Slot Interchanger (2)

• How does a TSI system work?
• Data is written to the speech store cyclically as
  it comes in (i.e. sequentially, one time slot at
  a time).
• Path set-up control signalling tells the SAM to
  store the name of the input time slot in the
  appropriate location corresponding to the output
  time slot it must be switched to.
• For example, if input time slot 7 is to be
  switched to output time slot 15, then location 15
  of the SAM will store the number 7.
• Data is read a-cyclically from the speech store
  in the order of the output time slots as stored
  in the SAM.
• Note that this means there could be a delay of up
  to nearly a full frame.

36
Time Slot Interchanger (3)
Data In (cyclic frame timeslot order)
Speech Store Stores the data of time slot x in
location x.
Timing
Write Address Counter
Timing
Read Address Counter
Control Signalling
SAM Stores the name of the input time slot being
switched to output time slot y. i.e. In output
time slot y, which speech store location do I
read?
37
Time Multiplexed Space Switch

• A TMSS is a space switch (with multiple physical
  inputs and outputs) that is potentially
  reconfigured entirely in every time slot of each
  frame.
• For instance, say we have 3 time slots on each of
  4 physical inlets and 4 physical outlets (also
  called I/P highways and O/P highways)

38
Time Multiplexed Space Switch (2)

• How does a TMSS system work?
• A memory structure called cross-point address
  memory (XAM) is used to control switching.
• XAM is a RAM with capacity to store a word for
  each time slot, each word being a number
  identifying a specific physical input to connect
  to during each time slot.
• Control signalling tells the XAM to store the
  name of the physical input in the appropriate
  time slot location.
• For example, if input 6 must be connected to
  output 9 during time slot 7, the the XAM for
  output 9 will store the number 6 in location 7.
• The space switch is rapidly reconfigured at each
  time slot to affect the proper connections.
• Note that data is switched across physical
  inputs/outputs, but not across time slots.

39
Time Multiplexed Space Switch (3)

• Column Oriented Control Who do I get it from?

Each XAM stores the name of the I/P to which its
O/P is connected to in each time slot.
Example To switch I/P 2 to O/P 4 in time slot
18, then XAM 4 stores the value 2 in location
18.
XAM 1
XAM 2
XAM 3
XAM 4
40
Time Multiplexed Space Switch (4)

• Row Oriented Control Who do I give it to?

Each XAM stores the name of the O/P to which its
I/P is connected to in each time slot.
XAM 1
XAM 2
XAM 3
Example To switch I/P 2 to O/P 4 in time slot
18, then XAM 2 stores the value 4 in location
18.
XAM 4
41
Time-Space-Time Switching beam me up Scotty -)
Space Switch
Physical inputs are connected to physical outputs
but data does not cross time slots.
Time Switch
TSI
Data is switched between time slots but remains
on the same physical connection.
A
B
C
D
A
B
C
D
B
D
A
C
B
D
A
C
Time-Space-Time Switch
TST
Data is switched between time slots and physical
connections.
42
Time-Space-Time Switching (2)

• Time-Space-Time switching is when data is
  switched across time slots and physical
  connections.
• Affected by a combination of TSI and TMSS.

43
Time-Space-Time Switching (3)

• What is the space division equivalent of a TST
  switch?

44
Time-Space-Time Switching (4)

• How does a time-space-time switch work?
• First, we find a time slot that is free from the
  input TSI to the TMSS and from the TMSS to the
  output TSI we wish to connect to.
• Next, switch the input channels time slot in
  question to the free time slot.
• Then at the TMSS, connect the proper input line
  to the proper output line during free time slot.
• Finally, at the output lines TSI, switch the
  free time slot to the time slot we wish to switch
  to.

TSI
TMSS
TSI
Input
Output
45
Multiplexing TSI Stages

• Multiplexing will increase the number of time
  slots into a TSI.
• Example

Speech Store 192 Bytes
81
SAM 192 Slots - each slot big enough to store
number as big as 192
81
46
Multiplexing TSI Stages (2)

• What benefits do we get by multiplexing?

Smaller P(B)